There is a vast gender imbalance in Software Engineering, with few women entering the specialisation. And the entry rates have remained the same over the years, despite attempts to encourage women. One potential reason is that children start to form opinions early in life (before ten years of age), which then influence the subjects they choose in schools, limiting their choices for university.

In this project, you will enhance and evaluate a physical programming environment designed for young children. In particular, we are looking at how the environment can improve girls’ perceptions of STEM and help them form positive opinions of the topic. As part of the project, you will develop a workshop for primary school-aged children and trial it in schools.

This project is a collaboration with the School of Design in CAI.

Location: Radio Systems 405.736

Children are often fascinated with robots and want to control (program) them. Robots also offer a pathway to teaching children computational thinking, an essential skill in our modern society. However, most programming environments are screen-based, requiring either a mouse and keyboard or a touch screen. These requirements can be difficult for young children, adding complexity and stress to the tasks. This, in turn, can lead to a negative perception of programming for children.

In this project, you will explore how computer vision and AI can enhance a physical coding environment as an alternative approach to programming robots. The project will build on previous projects to provide better physical blocks (tangibles). As part of the project, you will also evaluate the tangibles with young (primary school-aged) children.

Location: Radio Systems 405.736

Ethics is becoming increasingly important in society. But despite its importance, it can be a complex subject to teach. Students often have preconceived ideas and are not aware of the consequences of these views. Board games offer a fun, low-risk way to explore the consequences of our actions. Students are free to explore other viewpoints and see the results of their choices.

This project will look at designing and evaluating a board game for teaching ethics in a robo-ethics course at the university. You can choose the aspects of ethics you are interested in and design the game around them. The game’s requirements: the players should have fun while building their understanding of robo-ethics.

This project builds on a previous iteration of this from last year. It is a collaboration with the School of Design in CAI.

Location: Radio Systems 405.736

Trying to understand a large, unknown computer codebase is a challenging task. Currently, developers use Integrated Development Environments (IDEs) to navigate through a codebase while trying to connect different parts of the system. While the tools in an IDE can help, they still rely on people being able to build a mental map of the code. Another approach is to use people's spatial abilities to help with this process. A large screen code editor would allow developers to position code in ways that help them see paths between different components.

In this project, you will build a large-screen code editor with existing open-source components. Programmers should be able to import programs and files and rearrange them. In addition, you will evaluate the editor to determine whether it is helpful for understanding code.

Location: Radio Systems 405.736

The Formula SAE competition is introducing an autonomous vehicle category to the control of the Formula SAE cars. To meet this challenge we are seeking to develop a fully autonomous control system that can enable the car to learn to drive itself through Reinforcement Learning. Reinforcement Learning is a machine learning approach that seeks to enable a robot to learn to operate based on its own interaction with the environment without human input or control.

Work last year has started to develop a means of enabling a Formula SAE car to autonomously navigate through Reinforcement Learning based control in simulation with validation experiments on a Turtlebot 2 platform in the real world - this work can be seen here (https://www.youtube.com/watch?v=TYCrK8-dnqE). This project will extend this work to operate on an F1Tenth (https://f1tenth.org/) race car to navigate a "race track" in the lab using additional sensing and control modalities.

This project will require strong programming skills, specifically in python, and will require the students to frequently come into the robotics lab to work on the project. This project is not suitable for remote development due to the requirement of working with and testing on the physical vehicle. Prior experience with Pytorch or ROS (1 or 2) would be beneficial but not required as we will teach you those tools as part of the project.

The project forms a part of the SFTI Rangatahi Robotics project - https://www.sftichallenge.govt.nz/news/rangatahi-mission-lab/ and will work closely with the existing team in the CARES (https://cares.blogs.auckland.ac.nz/) group working in this space. We will be seeking to publish the results of this project at an international conference.

Location: Robotics 405.652

This project is a cross-departmental project seeking to develop a Reinforcement Learning (RL) based control system for the grasping of objects. RL is a part of Artificial Intelligence (AI) which seeks to enable robots to autonomously learn to perform tasks without human interaction or guidance. Through RL a robot learns to perform actions based on observations of its environment through its sensors and feedback in the form of “rewards” for taking good actions, aiming to maximize the overall reward it receives while carrying out a given task.

This project extends upon work completed last year that enabled a two-fingered robotic gripper to manipulate objects through the state-of-the-art in RL - this work can be seen here https://www.youtube.com/watch?v=2QBPtrrYIrY and https://www.youtube.com/watch?v=xkack0nEHlM.

This project will extend this work to a new three-fingered gripper (https://github.com/UoA-CARES/Gripper-Code) which will increase the complexity of the manipulation tasks the system can learn. Furthermore, we will apply novel RL techniques to both robotic grippers to explore further improvements to the learning system itself.

The physical gripper and experimental environment necessary for conducting these experiments have been constructed already - the teams will focus on the development and evaluation of the novel learning algorithms.

This project will require strong programming skills, specifically in python, and will require the students to frequently come into the robotics lab to work on the project. This project is not suitable for remote development due to the requirement of working with and testing on the physical grippers. Prior experience with Pytorch or ROS (1 or 2) would be beneficial but not required as we will teach you those tools as part of the project.

The project forms a part of the SFTI Rangatahi Robotics project - https://www.sftichallenge.govt.nz/news/rangatahi-mission-lab/ and will work closely with the existing team in the CARES (https://cares.blogs.auckland.ac.nz/) group working in this space. We will be seeking to publish the results of this project at an international conference.

Location: Robotics 405.652

The electrification of transport and development of more efficient renewable energy systems has pushed power electronics into cryogenic environments where superior performance and increase in power density has been observed. Being considered a viable method of reducing our carbon emissions and meeting climate goals, the desire to integrate cryogenic power electronics into such systems is growing.

GaN HEMTs have been proven to provide performance enhancements over traditional Si MOSFETs, for instance in power converters and motor drivers operated under cryogenic conditions. However, ancillary circuitry, for example the gate drivers, protection ICs, housekeeping power supplies, isolators, sensors, etc. required to implement a fully cryogenic power converter or a motor driver, fails at low temperatures. In addition, the drivers, protection ICs and control ICs developed for Si MOSFETs are not directly applicable to more recently introduced GaN HEMT devices.

Location: Green Electronics 405.712

A phase shift full-bridge converter can be used as part of an uninterrupted power supply (UPS) to charge supercapacitors employed to maintain the output of the UPS in the event of mains supply failure or sudden variation. The control of this phase shift full-bridge converter can be challenging as the supercapacitors need to be charged from 0V to its designed charged voltage. Under these strenuous operating conditions, the controller should be designed to be stable while also ensuring efficient operation of the converter. Thus, it is preferable to implement a digital controller to control a phase shift full-bridge converter that is used to charge supercapacitors used in a UPS.

The primary goal of this research project is to develop a digital controller that can synthesise optimum modulation for a phase shift full-bridge converter charging supercapacitors. After conducting a detailed literature research to understand control strategies, the research will first develop a simulation model of a phase shift full-bridge converter that can accurately model losses. Using this simulation model, performance of different controller implementations will be studied to select the most suitable control strategy. The research will look at varying the operating frequency and dead-time to reduce the losses. Finally, a prototype system, with specifications listed below, will be built to test and validate the research findings:

This work is completed in collaboration with ABB Napier.

Location: Power Electronics 405.614

Automatic speech recognition (ASR) can be very helpful for speakers who suffer from dysarthria, a neurological disability that damages the control of motor speech articulators. ASR can act as a medium to not only understand the impaired speech but also to talk on their behalf and enable them to have a better social and digital life. Nevertheless, normal speech recognition systems have not been able to understand dysarthric speech, leaving speech impaired individuals not being able to utilize ASR technologies technologies that could be life changing for them.

In this project you will investigate the latest advacments in speech recongtion using Transformer and Attention networks Deep Learning algorithms, and will build an ASR system. You will also leverage transfer learning and data augmentation to overcome the existing data scarcity challenges.

You can learn more about the project by reading this paper:

https://ieeexplore.ieee.org/document/9419963

https://ieeexplore.ieee.org/document/8462506

Location: HASEL 405.662

Introduction of LED light sources to commercial lighting systems allows significant energy savings. LEDs with lenses are becoming very popular. The lenses introduce various advantages and disadvantages, which will be investigated.

Location: Green Electronics 405.712

Challenges with food production are driving the growth of smart farms - indoor environments where climate and nutrients can be controlled to maximise efficiency and production. The greenhouse can promote products, increase stability and security, minimise production risk, control diseases and pests, and improve resource utilisation. Inconsistency and delay in identifying plant diseases cause productivity loss, although this can be avoided with constant observation. In this project, our focus is on detecting plant disease from its early stages.

The student will collect multispectral/NIR data from plants and annotate them. Bioforce Ltd sponsored this project and students should have some field trips to the company for data collection. Students will develop a deep learning-based computer vision model to detect diseases on leaves and stems. The research aim will be to extend the computer vision models to handle different kinds of diseases. This project will primarily be done using Python and/or C++ within a Robotic Operating System (ROS) framework.

In some cases, existing python projects exist from the CARES and MaarTech repositories, and these codebases may be used for this research.

Location: Robotics 405.652

There is a growing demand for skilled vine pruners to maintain the quality of vines for quality wine production. The challenge is providing an effective training process to produce enough workers to meet this demand. Prior work has developed a tech demo of a vine training tool based in VR using an Oculus Quest - https://www.youtube.com/watch?v=JYZHwr2fELc

This project will extend this tech demo into a fully-fledged training system and investigate how this tool could be gamified to help improve the quality of training and the workers' overall performance in the real world. A usability study will determine the tool's effectiveness in training people to prune vines.

This work is part of the CARES human-assist (MaaraTech) project funded by MBIE - https://cares.blogs.auckland.ac.nz/research/robots-in-agriculture/data-informed-decision-making-and-automation-in-orchards-and-vineyards/

Location: Radio Systems 405.736

Fault tolerance is an important property of any space mission. In large, expensive satellites, redundant hardware systems are used to improve reliability of computing in space to ensure their success. For smaller, cheaper satellites, like CubeSats, redundant hardware is often not an option. Nevertheless, it is of vital interest to make all satellite missions as reliable as possible. In this project we will investigate if and how we can make computing in CubeSats more fault tolerant. The focus will be on scheduling techniques for the various on board tasks of typical systems and payloads found in CubeSats. After identifying common operating systems and software stacks of CubeSats in an extensive literature review, we will identify which scheduling and computing strategies can improve fault tolerance. Existing scheduling techniques will be adopted to this scenario and we will fill gaps with novel approaches and algorithms.

Location: Radio Systems 405.736

Most satellites in Earth's orbit gather data from sensors which is then transmitted to their ground station. This can be simple sensor readings, but often involves considerable amounts of data, for example when images are taken. To reduce the amount of data to transfer, (pre-)processing of the data can already happen in the satellite. This can be done to reduce the amount of data, e.g. by compressing the data, or to carry out the processing of the data for the intended application, e.g. detecting objects in images. In either case, this means that the satellite needs to posses some considerable computing capabilities. At the same time, all satellites need to function with a limited power supply. This is especially true for small form-factor satellites like CubeSats. Hence, the necessary computing power needs to be achieved with as little power as possible. In this project we will investigate the design and implementation of power efficient algorithms for the application in small Satellite. Our focus will be on algorithms for imaging using techniques like Synthetic-aperture radar (SAR), phased array antennas or interferometry in radio astronomy, all of which are algorithmically related. After reviewing the literature, we will identify and select algorithm(s) to target for the project. These algorithms will then be designed and implemented in a development kit (NVIDIA Jetson family), that focuses on low-power high performance computing and provide realistic restrictions as one could find in a CubeSat computing system. The design and implementation of the algorithms will efficiently employ the available heterogeneous computing resources such as CPU, GPU and acceleration extensions.

Location: Radio Systems 405.736

Parallel computing has become enormously important especially in machine learning and neural networks. Often graphics cards (GPUs) are used to accelerate the execution of an otherwise long running program, e.g. the training of neural networks. When this is done, data in the main memory of the processor (CPU) needs to be transferred to and from the GPU. In order to get the best possible solution, these communication costs need to be considered when scheduling a program onto the CPU-GPU system. Modern scheduling algorithms consider this form of heterogeneity and try to model the communication costs. However, the models are still lacking, considering costs to simplistically. Often identical communication is counted multiple times, even though it is necessary only once. In this project you will investigate different scheduling algorithms that are applicable to modern GPU-CPU system, varying the system model and the complexity. We aim for general solutions, but will target workloads found for the training of neural networks. The algorithms will then be implemented and measured in an experimental study. The outcome will be an algorithm library to be integrated into existing research work of the Parallel and Reconfigurable Computing Lab.

Location: Radio Systems 405.736

Domain-specific languages (DSLs) for image processing have successfully allowed the exploration of high-performance algorithms on modern CPU and GPU systems. Techniques such as decoupling algorithms from their schedules create opportunities to introduce new computing platforms to DSLs using existing algorithms. Field programmable gate arrays (FPGAs) provide a platform for high-performance and low-power computation as reconfigurable accelerators. Image processing DSLs have used high-level synthesis (HLS) as an output that can be synthesised for an FPGA. This process leaves room for performance optimisation as tools that target HLS languages do not have insight into the implementation details of the target FPGA. This project will investigate the performance improvements that can be gained from the HLS compiler and synthesiser reports informing the DSL of the implemented circuit. The relevant information that can be used from the compiler reports to inform performance optimisation must be determined. This information can then be integrated into the DSL for performing optimisations when lowering to the HLS target. This project aims to determine the relevant information from HLS and synthesisers that DSLs can use when optimising and lowering, and the scale of the performance and resource utilisation improvement that can be gained from such information.

Location: Radio Systems 405.736

Wireless power transfer technology are used in places where the energy source and the load are far apart, such as drones and microsatellites. Energy efficiency is one of the critical considerations in the transmission process. Laser has the advantages of strong directionality and high energy concentration. As an energy carrier, the laser provides energy to the load, but there will be loss incurred throughout this process. This project aims to measure the energy loss at each stage from the energy source to the load and calculate the efficiency. The stages include the driver, emitter, receiver, and regulator that power the load. Possible electronic improvements will also be proposed.

Location: Green Electronics 405.712

There is a growing demand in the green energy environment for energy storage systems and one

potential solution to this is PEM hydrogen electrolysis cells. Currently, this technology is quite novel

and requires further research into both improving efficiency and lifespan. One new development is an

input current waveform that decreases the cell voltage thus improving efficiency.

This project aims to both develop a power conditioning system to supply the optimal input current

waveform and a control scheme to find and keep the electrolysis cell at peak efficiency. In addition to

this, the project aims to examine the effects of this current waveform on the degradation of the

electrolysis cell. The project will then aim to account for any change in degradation in the control

scheme by periodically checking the efficiency of the cell.

This project will also examine the use of this system with a variable renewable source by mimicking

the output of a photovoltaic cell.

Location: Power Electronics 405.614

Thermal analysis is often neglected in the design of power electronics. Thermal analysis allows us to predict and prevent failures by providing a quantitative understanding of the temperature rise in a system under various operating conditions. Knowledge of the electrical and thermal properties can then be used to optimize the performance of electronic devices and systems.

Existing literature for the thermal analysis of electronics primarily uses thermal resistance calculations to find steady-state temperatures. Thermal resistance calculations largely neglect the transient response and environmental effects, e.g. enclosures. Computational Fluid Dynamics (CFD) is a technique to accurately model and predict temperatures accounting for these necessary considerations. By using CFD codes such as Ansys lcepak, we aim to identify and address key heat transfer mechanisms in these devices and to develop strategies for improving thermal management.

One such technology wherein the benefits of thermal analysis could be exploited is the power electronics of Inductive Power Transfer (IPT) systems. In such systems, capacitors are considered to be, thermally, some of the most fragile components as their characteristics change with temperature. Therefore, to improve reliability and extend the lifespan of the system while reducing costs and decreasing volume, capacitor boards have been chosen as the focus of this project.

This project aims to model the thermal characteristics of capacitor boards to accurately predict their thermal behaviour. This characterization can then be used to optimize these boards, considering factors such as placement, copper pours and materials.

Location: Power Electronics 405.614

Collecting and labelling large data sets for machine learning (ML) is difficult and costly when starting a new project. In a previous Part IV project, we proved the feasibility of using synthetic data for detecting pallets in photos.

This prior PTIV project provides:

Crown is interested in the following open questions:

Location: Embedded Systems 405.760

This project aims to bridge the gap between University education and the skills required in technical interviews in the software industry. A large majority of software companies include some components of algorithm and data structure questions as part of their hiring process. While the theoretical foundations for such questions are taught in some University courses, the style of this content doesn't match well with the practical-focused questions encountered in technical interviews.

We propose to design, develop, and evaluate a prototype web application aiming to provide students with interactive examples, questions and explanations from scratch, so that they are confident in solving interview-style questions.

Research Components:

Location: MDLS 405.559

With sufficient understanding of the architectural design for a software application, software engineers are enabled to develop quicker while also able to apply better coding practices in line with existing software design tactics and principles. However, code evolves frequently without accompanying changes in documentation or changed to a poor standard, because documentation and keeping diagrams up-to-date is often seen as less worthwhile. This often hinders the ability for newcomers to learn and build understanding, especially in aging codebases. We believe there is a way to improve this through the introduction of an automatic architectural diagram generator that generates diagrams that evolve with the codebase.

Our research proposal aims to first understand what holistic attributes support intuitive and effective presentation formats as observed from current best practices, and what shortcomings they possess when it comes to building a user's understanding of a software's architecture. We then desire to propose a new standard of architecture representation that provides greater value in comparison.

We then want to confirm whether it is possible to automatically generate architecturally significant diagrams from a source codebase that conforms to this standard. In particular, we're interested in what indicators in code would aid humans (and hence, a computer program) to make specific visual distinctions in these diagrams. If we observe this to be theoretically possible, we want to develop a proof of concept that supports a limited set of languages (primarily Java) and a subset of architecturally significant design patterns. We also wish to explore what additions (in the form of comments, annotations, custom libraries etc.) are needed in the code base to support diagram creation if the code itself is not enough.

Location: MDLS 405.559

New Zealand Māori is the indigenous language and one of the official languages of Aotearoa New Zealand; Cook Islands Māori is the indigenous and official language of Cook Islands. Both languages have similarities, but they are also quite different from each other. Both languages are under-resourced in speech technology, with very few speech recognition or speech synthesis systems developed for the two languages. Recent efforts to develop speech technology for both languages are underway, thereby ensuring that both languages can digitally thrive. It is essential to have languages represented in digital technology in the present digital age. Otherwise, the usage of the language by the younger generation may reduce, thereby reducing the overall use of the language. With speech technology research concentrating on a few well-studied languages (99 out of 7097 world languages) like American-English, speech synthesis and recognition models are trained predominantly on these languages. Hence, speech technology development for our languages should come from us! Previous Part 4 projects along with the Speech Research @ UoA research group (https://speechresearch.auckland.ac.nz/) have developed text to speech synthesis systems for New Zealand Māori and Cook Islands Māori. This project aims to start with an existing text to speech synthesis model in a well-resourced language and fine-tune the model to that of New Zealand Māori. Using the knowledge gained, we will also develop the first text to speech synthesis system for Cook Island Māori. The project will involve:

1. Familiarisation of text to speech synthesis system development.

2. Use existing resources in the two languages to fine tune a text to speech synthesis model.

4. Develop a New Zealand Māori and Cook Island Māori text to speech synthesis system.

Being an expert in Māori is not essential for this project; instead, an awareness and appreciation for languages, languages differences and the willingness to learn. We will be working with Sally Ake Nicholas from Māori Studies, who will guide us on the linguistic aspects of the two languages.

Co-supervisors: Dr Catherine Watson (Electrical, Computer Systems and Software Engineering), Dr. Justine Hui, Mechanical Engineering

Location: Control Systems 405.722

Emotions are felt. But can they be detected and visually represented as well? Yes, there are two or three dimensional models that help us visualise our emotions. An example of such a model is the Russel's circumplex model of emotions that depicts emotions in a two-dimensional space of valence and arousal dimensions. Here, valence indicates how positive or negative an emotions is. And arousal indicates the if the emotion was expressed using high energy or low energy. Previous research at UoA (by Speech Research Group @ UoA https://speechresearch.auckland.ac.nz/) has developed speech emotion recognition algorithms and also some approaches to visualise emotions. Isn't that cool to be able to "see" the emotions we feel? Lets take that a bit further then!

We are all used to technology that speaks to us and recognises what we say, such as Siri, Alexa, Google Assistant, and similar human-computer interaction technology. With the demand for contactless interaction increasing globally, systems that can interact with users via speech are becoming more relevant than ever. Also, speech technology is trying to recognise the emotions in human speech to respond to human users effectively. This is where Speech Emotion Recognition (SER) in speech technology becomes used. But this technology development is often concentrated on a few well-studied languages (99 out of 4097 world languages) like American English and Mandarin Chinese. The languages spoken by users of speech technologies are not restricted to these well-studied languages. As a result, speech technology may not correctly recognise the emotions of speakers of less-studied languages. One example of a less-studied language is New Zealand English, where all the technology development for Aotearoa New Zealand must come from Aotearoa itself.

The aim of this research project is to integrate speech emotion recognition to approaches that visualise emotions. This means we will be able to not just recognise people’s emotions but also be able to display them in an intuitive manner. We will also aim to develop this specifically for New Zealnd English accent. The pre-requisites of this project are Python/Java, developing GUIs using JavaFX/PyQT, developing. The project will involve

1. Conducting a systematic literature review of speech emotion recognition approaches and emotion annotation, visualisation techniques.

2. Using existing emotional speech databases to train a Speech Emotion Recognition model and fine-tune it to New Zealand English.

3. Design a GUI/Web interface to visualise and annotate emotions and evaluate its design.

4. Test the system with speech samples unseen speech recognition model.

Co-supervisor: Felix Marattukalam

Location: Control Systems 405.722

There is an ongoing demand for contactless deliveries and more due to the recent pandemic. This has spurred an increase in using drones for different purposes, especially for contactless delivery. Computer vision algorithms, especially machine learning (ML) based algorithms are used to control drones. Following the recent AI trends, image recognition technologies for object recognitions (such as faces) can be used to make drone operations more autonomous, secure, and safer. Image recognition processes demand high amounts of computing power, and drones with their limited resources may not use the conventional computers. FPGA devices provide opportunities for hardware/software co-design to investigate the design trade-offs and support dynamic reconfiguration where needed to adapt to the environmental changes and limited available resources.

In this project, machine learning based implementations for face recogntion will be explored. An FPGA implementation for will also be considered as an edge device using both the configurable logic resources and hard-core processors to optimise the performance and accuracy of the system.

Location: Embedded Systems 405.760

Security vulnerabilities of network connected embedded systems, which are deployed in Internet of Things (IoT) devices have become significantly important especially in safety critical applications such as medical devices and autonomous vehicles. Due to continuous changes of security threats, employing a run-time potential security threat detection is essential. This can be considered at different components of the multi-processor systems architecture. Processing elements are the main components of such systems and identifying the security threats efficiently depends on their complexity. In this research, a softcore configurable processor based on RISC-V instruction set architecture (which is an open-source instruction set architecture) is extended to support information flow tracking at run-time. The effectiveness of the proposed architecture extension and its performance overhead are evaluated.

Location: Embedded Systems 405.760

Multi-Processor Systems on Chip (MPSoC) are the most efficient implementation platform for safety critical embedded applications when the performance, energy efficiency and time predictability are the main design concerns. Due to the network connection of embedded devices addressing security issues have become an important design issue. Recent studies indicated the security vulnerabilities of different components of an MPSoC including modern processor microarchitecture, the memory systems and the interconnection network which, may result in side-channel attacks. Side channel attacks refer to retrieving security critical information from the device due to the weakness of physical implementation of the system.

The aim of this project is to investigate the potential side channel attack possibilities for FPGA-based MPSoC platforms considering either a case study application or some standard benchmarks applications. More detialed investigations should be related to the Network on Chip interconnection in MPSoC.

Location: Embedded Systems 405.760

A key concern in the modern world is human activities leading to environmental changes, which could lead to a lower quality of life. The development of technologies that allow us to enjoy our planet's resources with greater efficiency and less damaging impacts seems to address some of this concern.

Aquaculture is the idea of farming aquatic resources - which brings with it the promise of a much more controlled and responsible approach to managing the ocean's resources. While aquaculture is not a new idea (it is maybe 4000 years old), its main drawback is the challenge of working underwater.

The Centre for Automation and Robotic Engineering Science (CARES) is interested in developing underwater robotic systems to help automate typical aquaculture activities to make it far more viable as a strategy for preserving New Zealand's aquatic resources.

This project focuses on a critical obstacle in developing a "vision" system for such a robot, namely the ability to interpret correctly images captured in turbulent water. Turbulent water is water in which the speed of the fluid, at any point, continuously changes in both magnitude and direction. Images acquired in such conditions can be pretty distorted.

Your research will explore the utility of a machine-learning approach to try and "correct" the effects of underwater turbulence. As a starting point, looking at Learning to see through turbulent water may be worthwhile.

Location: Power Systems 405.628

Pollution from wood burners has serious health implications for the general population. Typically, monitoring the level of airborne particulate matter in these areas often requires making inferences about missing or corrupted readings. Air Quality inference in these cases often poses critical challenges. These factors can evolve over time, changing the distribution of data. Methods incorporating environmental factors to capture these changes, e.g., weather, traffic and points of interest, have found success in urban environments. Such changes in the distribution of data are known as concept drift. We will investigate five air quality studies in New Zealand's rural towns. We will develop a transfer learning for the data stream whereby we reuse a pre-trained air pollution inference model from a town as the starting point for an air pollution inference model on the town. Transfer learning is a core machine learning paradigm that is able to address learning problems in small data settings. Transfer learning is founded on the belief that multiple, differing tasks are distinguished by a small number of task-specific parameters but often share a common low-dimensional representation. We will investigate current data stream models alongside machine learning models encapsulating the transfer learning paradigm.

Location: Embedded Systems 405.760

Environmental damage due to human activity has the potential to destroy otherwise sustainable resources and could lead to environmental changes that could substantially affect our quality of life. Therefore there is a global focus on developing technologies that allow us to retain the current benefits we get from the environment in a much less impactful way.

Aquaculture is the idea of farming aquatic resources - which brings with it the promise of a much more controlled and responsible approach to managing the ocean's resources. While aquaculture is not a new idea (it is maybe 4000 years old), its main drawback is the challenge of working underwater.

The Centre for Automation and Robotic Engineering Science (CARES) is interested in developing underwater robotic systems to help automate typical aquaculture activities to make it far more viable as a strategy for preserving New Zealand's aquatic resources.

This project focuses on a critical obstacle in developing a "vision" system for such a robot, namely the ability to correctly interpret images captured in turbid water.

Turbid water is a commonly encountered "real-world" phenomenon in which a body of water has a high density of particles that are suspended or dissolved. Imaging under turbid water yields murky images in which elements are only partially visible due to effects such as light scattering. Images that have been captured in turbid water are far more difficult to work with from a computer vision standpoint, because many of the traditional assumptions that computer vision algorithms use, cannot be applied to these images.

This is a very research-orientated project, in which the main research question is to determine the extent to which modern machine-learning strategies are useful in enhancing images captured in turbid water for computer vision. An example of a potential solution to this problem could be https://openaccess.thecvf.com/content/ACCV2022/papers/Tang_AutoEnhancer_Transformer_on_U-Net_Architecture_search_for_Underwater_Image_Enhancement_ACCV_2022_paper.pdf

Location: Embedded Systems 405.760

Gastroparesis is a disease of the gut, which leads to slow down of the gastric pacemaking waves. The slowdown of the gut waves leads to indigestion, bloating, and can be life altering. Gastric pacemakers can help alleviate the problem. However, no gastric pacemaker is currently (FDA) approved for medical application.

We have developed closed-loop gastric pacemakers targeting brady-gastria (slow-down). Our pacemaker has been developed in Simulink/Matlab and we have also performed very preliminary animal experiments. Although the simulation results are good. The animal experiments are not very convincing. The algorithm in the pacemaker is unable to correctly detect the activation of the human gut many a times. This is mainly because the pacemaker algorithm is static, while the human/animal gut beats at different frequencies depending upon the status (with food, sleeping, etc). We want to develop a machine learning technique to correctly detect the gut activation. If the gut is correctly activated then there is no need to apply a manual pace. However, if the gut does not beat correctly then the pacemaker needs to do its job of applying the pace.

Location: Embedded Systems 405.760

Autism (Takiwātanga) is a lifelong developmental condition affecting the social and communication skills of one in 59 (>93,000) New Zealanders. Early identification is important but is currently very delayed in New Zealand. Particularly, the current mean age of diagnosis in NZ is between 6-7 years - usually 2-3 years after families/carers have expressed concerns. The optimal window for delivering treatment to children with autism is between 2-3 years, necessitating identification before two years of age. Early diagnosis is also important for neuroplasticity-related language acquisition, which is greater when children are younger and prognostically more beneficial prior to them starting school at five years of age.

In this project, you will develop and investigate how inductive-transductive learning could be applied to utilize the limited labeled dataset we have collected.

You can read more here:

Inductive-Transductive Learning:

1. https://towardsdatascience.com/inductive-vs-transductive-learning-e608e786f7d

2. https://www.sciencedirect.com/science/article/pii/S0925231222007822

3.. https://academic.oup.com/bib/article/23/4/bbac207/6604996

ASD Detection:

1. https://link.springer.com/article/10.1007/s12559-020-09743-3

We will use Keras, Python, and Flutter in this project.

Location: HASEL 405.662

Power quality has been a major reszearch concern due to presence of various switching power electopnics devices and nonlinear loads. To alleviate the prob lems associated with power quality, it is essential to correctly identify these events. The proposed project will develop machine learning based algorithgms including deep learning for power quality identification. The students must be well conversanmt wigth MATLAB; although knowledge of Python may be better.

Location: Control Systems 405.722

Problem Statement

The controversial events surrounding the GameStop stock, driven by the WallStreetBets community on Reddit, has brought to light the potential impact of social media on financial markets and the potential consequences it can have on the real world. The use of social media indicators, such as sentiment and engagement, to predict stock/cryptocurrency performance has become a crucial area of research for investors, regulators, and market participants. This research project aims to investigate the correlation between social media indicators and stock market performance, specifically focusing on the GameStop case as a real-world example. The goal is to understand the extent to which social media data can be used to inform investment decisions and potentially improve returns for investors, while also assessing the implications for market efficiency and regulation.

Motivation

The price surge of GameStop under the influence of WallStreetBets subreddit has sparked interest in understanding the role of social media in stock market trading and its potential implications for market efficiency and regulation. Additionally, understanding the potential of using social media data to improve investment returns and obtain more market insights are largely desired by financial institutions and retail investors.

Key prior research efforts on the problem

Several studies have investigated the relationship between social media indicators, specifically from Reddit and WallStreetBets, and stock market performance. The publication "Analysing the Hidden Causal Interactions in Large-Scale Social Networks: A Case Study on GameStop" [1] has demonstrated a causal interaction between the Gamestop price surge and the WallStreetBets community. This study also found evidence of causal interactions between WallStreetBets and cryptocurrencies.

Other publications have also highlighted the strong correlation between Reddit sentiment and stock prices. For example, the study "Stock Price Prediction Leveraging Reddit: The Role of Trust Filter and Sliding Window"[2] found that using a trust filter and sliding window improves the accuracy of the prediction model. Additionally, the study found that neural network models tend to perform better than other models. Similarly, the study "Cryptocurrency ecosystems and social media environments: An empirical analysis through Hawkes’ models and natural language processing" [3] found that applying Hawkes models on market sentiment data yields reasonable accuracy in predicting stock prices.

Given the evidence from prior research that Reddit can have a significant impact on stock prices, there is a need for further investigation into how Reddit reflects on the market and the potential implications for investment strategies and market regulation.

Contributions

Location: MDLS 405.559

Increasing usage of renewable energy is a major focus worldwide. In spite of being abundant and freely available from nature, renewable sources like solar, wind individually have the intermittency problem. Microgrids are the solution to integrate the renewable energy source, energy storage system and local loads, which connected by power electronic devices. Hence, the design of an efficient, robust and optimal control method used in power electronic devices is an important part of the research of microgrids and green energy technologies. Thanks to the development of digital microprocessors, the increase in processing capabilities allow applications of the model predictive control. In this project, the photovoltaic and hybrid energy storage system will be controlled and coordinated based on state-of-the-art control methods like model predictive control, PID. The test-bed at Power Systems Lab at Dept of ECSE, UoA would be used to experimentally validate the concept, and extend it further.

Location: Power Systems 405.628

Cyber-physical systems, such as autonomous vehicles, use interactive machine learning modules for decision making. Current approaches use a set of interacting neural networks, called neural network ensembles, to design such systems. These ensembles are realised using multithreading-based composition, which is not ideal for safety-critical systems. Hence, for safety-critical systems such as autonomous vehicles, there is a need for rigorous approaches.

This project seeks to develop an alternative approach by reusing neural network ensembles, which are designed in the Keras library. After the application has been developed and the networks are trained, a compiler (recently developed by us), will be used to generate automatic C code, where the neural networks will operate synchronously. Our approach, thus, brings model driven engineering to the AI space. This will facilitate automatic static analysis to determine the worst case timing behaviour of the application, which is important for autonomy.

The students will develop NN ensembles for autonomous vehicle application in Keras using deep neural networks. These will be then used to design an automated platooning system using the F1/10 racing vehicles (see: (1) http://www.madhurbehl.com/f110.html (2) https://www.youtube.com/watch?v=vgEyvazwrU8).

The prior work on this at UPenn (Ivanov et al. below), the car only supports uniform motion. Hence, it can only deal with static obstacles. Also, they only a single car is validated. We will develop an approach for a platoon of cars following each other and will support both static and dynamic obstacles.

References

Roop PS, Pearce H, Monadjem K. Synchronous neural networks for cyber-physical systems. In2018 16th ACM/IEEE International Conference on Formal Methods and Models for System Design (MEMOCODE) 2018 Oct 15 (pp. 1-10). IEEE.

Allen N, Raje Y, Ro JW, Roop P. A compositional approach for real-time machine learning. InProceedings of the 17th ACM-IEEE International Conference on Formal Methods and Models for System Design 2019 Oct 9 (pp. 1-5).

Yang X, Roop PS, Pearce H, Woo JR, A COMPOSITIONAL APPROACH USING KERAS FOR NEURAL NETWORKS IN REAL-TIME SYSTEMS, Design Automation and Test in Europe (DATE), March 2020, Grenoble, France.

R. Ivanov, J. Weimer, R. Alur, G.J. Pappas, and I. Lee. Case Study: Verifying the Safety of an Autonomous Racing Car with a Neural Network Controller, arXiv, 2019.

R. Ivanov, J. Weimer, R. Alur, G.J. Pappas, and I. Lee. Verisig: Verifying safety properties of hybrid systems with neural network controllers, 22nd ACM International Conference on Hybrid Systems: Computation and Control, 2019.

Location: Embedded Systems 405.760

Traditional wireless power transfer (WPT) systems using inductive power transfer (IPT) are designed for low coupling scenarios, usually between 0.1 and 0.3. In these cases, filters are designed to keep the coil driving current as sinusoidal as possible. The sizes of the magnetic systems required to deliver power can be decreased if the coupling factor can be increased up to 0.9.

The goal of this project is to investigate how high coupling factors will impact the total harmonic distortion of the currents using various tuning topologies (series-series, LCL-parallel, LCL-LCL).

Location: Power Electronics 405.614

Every developer knows that good quality code has meaningful identifiers. However it can be difficult to decide how meaningful an identifier is, and this judgement can be subjective. Ideally there would be some form of measurement that tells us how good an identifier is, but the subjective nature of evaluating meaningfulness makes this difficult. Nevertheless there are heuristics and proposals for forms of measurements that are referred to when discussing what makes an identifier meaningful. The goal of this project is to build a tool that, given some code, will provide a score for how "meaningful" each identifier in the code is. This will likely involve the use of Information Retrieval (IR) and Natural Language Processing (NLP) techniques, such as Latent Semantic Analysis to identify relevant topics, and part-of-speech analysis to analyse the make-up of identifier names. There are online facilities to help determine information about words, such as WordNet (https://wordnet.princeton.edu). WordNet has an API allowing for programmatic access (http://www.lexicadb.com/lxserver/wn_http.html), and there has been some previous work done (although quite old now) [1], [2]. The resulting tool will be demonstrated on source code. There are a number of possible datasets available, including submissions to programming assignments and open-source Java software. [1] "Leveraged Quality Assessment using Information Retrieval Techniques" Lawrie, H. Feild and D. Binkley, 14th IEEE International Conference on Program Comprehension (ICPC'06), Athens, Greece, 2006, pp. 149-158, doi: 10.1109/ICPC.2006.34. https://doi-org.ezproxy.auckland.ac.nz/10.1109/ICPC.2006.34 [2] "Improving Source Code Lexicon via Traceability and Information Retrieval" A. De Lucia, M. Di Penta and R. Oliveto, in IEEE Transactions on Software Engineering, vol. 37, no. 2, pp. 205-227, March-April 2011, doi: 10.1109/TSE.2010.89. http://ieeexplore-ieee-org.ezproxy.auckland.ac.nz/document/5601742

Location: MDLS 405.569

One aspect of "code quality" is the choice of names for identifiers. Guidelines have been proposed to help developers choose good names (e.g. [1]). For example, the Java Language Specification says "Names of class types should be descriptive nouns or noun phrases, not overly long, in mixed case with the first letter of each word capitalized." Some aspects of such guidelines are easy to check (e.g. "not overly long" or even "first letter of each word capitalized"). But some aspects refer to the fact that names are typically based on natural language characteristics, such as "parts of speech" (POS). POS refers to such things as nouns, verbs, past participle tense, an so on (eg. "Names of class types should be descriptive nouns"). The goal of this project is to investigate the use of POS guidelines, such as what guidelines exist and to what degree are they followed. This will require investigating POS "taggers" - tools that determine what POS a word in an identifier name has. Several taggers exist, but their accuracy for names found in source code can be variable because identifier names are often not grammatically correct (e.g. [2]). However it has been suggested that simple changes can improve tagger performance (e.g. [3]). There are several directions this project could take. Including: * Finding (and perhaps developing) POS-based identifier naming guidelines and applying them to existing code * Improving the quality of POS taggers, which will require assessing existing taggers * Examining how identifier names are used based on POS * Any combination of the above. The resulting tool will be demonstrated on source code. There are a number of possible datasets available, including submissions to programming assignments and open-source Java software. NOTE: To be successful, this project requires a good understanding of "grammar", or at least willingness to gain a good understanding quickly. [1] On the Naming of Methods: A Survey of Professional Developers. R. Alsuhaibani, C. Newman, M. Decker, M. Collard and J. Maletic, 2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE), Madrid, ES, 2021, pp. 587-599, doi: 10.1109/ICSE43902.2021.00061. https://ieeexplore-ieee-org.ezproxy.auckland.ac.nz/document/9401967 [2] Part of Speech Tagging Java Method Names. W. Olney, E. Hill, C. Thurber and B. Lemma, 2016 IEEE International Conference on Software Maintenance and Evolution (ICSME), Raleigh, NC, USA, 2016, pp. 483-487, doi: 10.1109/ICSME.2016.80. https://doi-org.ezproxy.auckland.ac.nz/10.1109/ICSME.2016.80 [3] Improving identifier informativeness using part of speech information. Dave Binkley, Matthew Hearn, and Dawn Lawrie. 2011. In Proceedings of the 8th Working Conference on Mining Software Repositories (MSR '11). Association for Computing Machinery, New York, NY, USA, 203–206. http://doi.org.ezproxy.auckland.ac.nz/10.1145/1985441.1985471

Location: MDLS 405.569

This project is to build and evaluate a tool that supports "Questions about Learner's Code" (QLC). Reflection is an important part of learning. Looking back at an answer giving for an assignment or test, and thinking about how the answer was developed, and examining it for weaknesses can help determine the level of understanding of the course content the assignment or test was assessing. This is equally true of code submissions. The idea is that it is useful to be able to have students demonstrate their understanding of their own code. This might take the form of an instructor meeting with the student and asking questions about the code. For example, being able to say what the scope of a variable is demonstrates understanding of the programming language being used. However this is expensive in time, and it is difficult to avoid delays. The best time to ask about code is just after it has been written. Having a tool that automatically creates questions about the code would allow students to immediately test their own understanding of what they have done, and so support learning. The goal of the project is to develop a tool that supports QLC for Java. One already exists for JavaScript ([1]), so this can be used as a source of inspiration. It will be expected that other question types will be developed. [1] Teemu Lehtinen, Lassi Haaranen, and Juho Leinonen. 2023. Automated Questionnaires About Students’ JavaScript Programs: Towards Gauging Novice Programming Processes. In Proceedings of the 25th Australasian Computing Education Conference (ACE '23). Association for Computing Machinery, New York, NY, USA, 49–58. https://doi-org.ezproxy.auckland.ac.nz/10.1145/3576123.3576129

Location: MDLS 405.559

Radio frequency (RF) wireless energy harvesting capitalizes on the conversion of RF waves abundant in the environment to drive low-power devices. The technology has a huge potential for the replacement of batteries, which contribute to a great amount of waste and pollution alongside their downsides of bulk and limited lifespan. The technology covers a wide range of applications from wireless sensor networks to biomedical devices to embedded systems. On a larger scale, with the intentional addition of a directed microwave source, this principle has the potential for long-range power transmission and the replacement of cables and batteries.

The primary goal of this project is to develop an RF energy harvesting system in an indoor environment. The system will include three main building blocks: an antenna, an impedance-matching network, and a rectifier or voltage multiplier. An energy storage device will be used for power management. There are numerous factors to consider in the design of such a system, including power conversion efficiency, manufacturability, operation distance, and size and sensitivity at different frequencies.

Location: Power Electronics 405.614

Ultraviolet (UV) lamps in the C band (with a wavelength range of 200-280 nm) are widely used to kill bacteria, viruses, and fungi for water treatment, air purification, and surface disinfection applications. This project is to investigate the feasibility of developing a contactless power supply for driving UVC lamps. The work involves a literature review of existing techniques, design of a new contactless power supply, circuit simulation, and practical verification.

Perspective students should have a very good background in electric circuit design. Successful candidates will work closely with postgraduate researchers and potentially be funded to continue Master or Ph.D studies after graduation.

Location: Power Electronics 405.614

This goal of the project is to design intelligent controllers for effective control of renewable energy systems with special emphasis on wind energy based energy systems. The efficacy of the designed controllers will be esstablished via OPAL-RT. The controller desugn should be generic and can be applicable. to control other types oif systems. The students must be well versed with MATLAB.

Location: Control Systems 405.722

With the likes of ChatGPT, Copilot---and others to come---what impact will this have on computing students learning fundamental programming concepts? This project will explore differences of code written by humans, compared to code produced by generative language models. This project will bring together a number of different technical skills in order to answer such questions.

Location: HASEL 405.662

Reflections play an important role in the learning process, helping students think critically about and grow from their experiences. As machine learning technologies make unprecedented progress, as demonstrated in the likes of ChatGPT, how can this be used to assist students be more reflective about their learning experiences? This project will involve exploring existng machine learning models that can help encourage this. The output of this project will be a tool incorporating such technologies, and evaluating their impact in the context of a range of courses.

Co-supervised with Allan Fowler (https://profiles.auckland.ac.nz/allan-fowler)

Location: MDLS 405.559

How do students collaborate together in group-based software projects? This project will involve developing a tool to investigate patterns of collaboration in Git-based repositories shared by a team of developers. What group dynamics can be observed? Do students work together on the same features, or are they working on mutually-exclusive features? Do they switch roles? Outcomes will incude a tool that can study a Git repository, and extract key patterns associated to team collaboration.

Co-supervised with Juho Leinonen (https://juholeinonen.com)

Location: Radio Systems 405.736

The ability to detect plagiarism in work submitted for tertiary assessment has always been sought after in universities. The move towards more digital means of teaching and assessment, driven largely by the COVID-19 pandemic, only exacerbates the need for quality plagiarism detection. Of note, the release of technologies such as ChatGPT and Copilot has rekindled a discussion in this field, with many academic institutions concerned that such tooling would result in submissions not reflective of student understanding of course material. Within this, there is the consideration of plagiarism in programming assessments, which existing tools such as MOSS (Measure of Software Similarity) seek to address. We believe there is much work to be done here and thus, we would like to centre our project on investigating additional technologies and techniques to improve the defence against plagiarism in university programming assessments.

Co-supervised with Juho Leinonen (https://juholeinonen.com)

Location: MDLS 405.569

Body Area Networks (BANs) are wireless networks in which the communications links involves physical channels that are in close proximity to or even within the human body. These networks are finding applications in a range of areas including disease monitoring, emergency communications in rescue environments and military/battlefield communications. The ‘body area’ environment is challenging for communication systems design due to the presence of a lossy, inhomogeneous medium (the body) and a requirement to keep the antennas as small as possible. This latter requirement is especially challenging since antennas which are substantially smaller than a wavelength can be difficult to realise. The aim of this project is to develop a small antenna for a body area network (involving both on-body and off-body wireless links) in the 2.4GHz ISM band (and other bands, if appropriate), and will involve the identification of appropriate antenna structures, modelling their behaviour using CST Studio Suite/Microwave Studio, and fabricating and measuring their actual performance using the test facilities provided in the Radio Systems and Telecommunications Laboratory.

Location: Radio Systems 405.736

The demand of electricity keeps increasing in this modern society and the behavior of customers vary greatly from time to time, city to city, type to type, etc. Generally, buildings are classified into residential, commercial and industrial. This study is aimed to distinguish the loads for various types of residential buildings and establish mathematical and machine learning models to represent and model the load profile of each type, and compare them. Modeling household energy consumption is the first step in exploring the possible demand response and load reduction opportunities under the smart grid initiative. Residential electricity load profiling includes the details on the electrical appliances, its energy requirement, and consumption pattern. The model is generated with a bottom-up load model. Extensive simulation will be performed, with the aim of creating a user-friendly software application to track the same.

We will use the following programming languages, frameworks and libraries during the project:

· Python

· Scikit Learn, Keras

· Software tool development via Flutter

Location: Power Systems 405.628

Modern 5G communication systems make use of the millimetre-wave spectrum, i.e. frequencies between 30 and 300 GHz. These systems require a range of different antenna types, including both directional antennas that radiate predominantly in one direction, and omnidirectional antennas that (nominally) radiate well in all directions. A good example of an omnidirectional antenna is the half-wave dipole, but these are challenging to fabricate at millimetre-wave frequencies due to their very small size and lack of a waveguide feed. Broadbeam antennas fall somewhere between these limits - they have a dominant radiation lobe but this lobe is much wider in angle than traditional directional antennas. These antennas may find application in providing millimetre wave coverage in an office when placed at some location on the office periphery (typically high up on one of the walls)

The aim of this project is to design a practical broadbeam millimetre wave antenna incorporating a rectangular waveguide feed by (i) identifying appropriate structures; and (ii) designing, modelling and optimising a range of candidate designs using CST Microwave Studio. If time permits one or more prototypes may be fabricated.

Location: Radio Systems 405.736

Machine learning is increasing being used in speech analysis for automatic speech feature extraction. It has immense appeal as many of the machine learning platforms are much easier to use than the traditional speech processing methods. When it comes to the recognition of speech machine learning has been used with great success. However the question is can machine learning techniques be used for accurate measurement of specific speech features? This project investigates the use of machine learning in the identification of an important speech feature, called a formant, and compares the results to traditional formant estimation methods.

Location: Control Systems 405.722

While spatial audio has often focussed on simulating sound sources separated from the receiver, the field of study concerning real-time simulation of self-generated sound (eg. the voice) in virtual acoustic environments is limited. This project will involve reproducing the system architecture of one such study (Arend et al., 2014) and evaluating its use as a practice tool for vocalists. This system would allow singers to binaurally experience the effects of various rooms on their voice in real time. The project would involve taking acoustic measurements, consulting vocal tutors at the School of Music, developing the signal processing system, and designing and conducting experiments investigating the system’s value for vocalists.

Location: Control Systems 405.722

This proposed research project will investigate and attempt to implement a noise cancelling algorithm in conjunction with a machine learning process in order to eliminate noise from noisy inputs and reconstruct audio such as speech without/minimal effect to clarity. The machine learning algorithm will use datasets such as set sentences from individuals to train the A.I.

Location: Control Systems 405.722

Tunable white lighting has been around for some time. Its primary goal is to create changeable lighting conditions in building where occupancy factors are very high (hospitals, retirement villages, schools). A typical controller of the tunable white luminaire receives messages related to the illuminance levels. In addition, it follows the astronomical clock settings. In this project we will be developing a controller monitoring natural light, and replicating similar indoor lighting conditions.

Location: Green Electronics 405.712

The proposed research project relates to the ability of a user to detect and avoid phishing scams through a gamified framework. It builds upon existing theoretical models and focuses on observing a measurable improvement in the user’s competency at detecting such attacks.

This will involve a literature review based around existing related studies then the development of a playable game for mobile platforms. The game will adapt a theoretical model from pre-existing research to increase user competency regarding phishing attacks.This model will then be extended as to observe “progression” i.e. a measurable improvement in the user’s present competency as compared to when they first used the game.

The game is designed to record analytics from the user’s playthrough in regards to their behavioural patterns when dealing with a possible phishing attack. These will be recorded and categorised over the course of their playthrough and the game will be able to produce a summarisation of their ability level and how it has changed.

The research will focus on the suitability of the game format for implementing observability of user progression. It could then, in further research, be used to empirically investigate the effectiveness of the game — and in turn model — at influencing user behaviour to better protect them against phishing attacks.

Location: Green Electronics 405.712

This project will develop a system to monitor the changes of plant growing. We have a working version that detects target, which is tomato, and measures its size and volume. May need to train Deep Neural Network (DNN) models to improve the accuracy of existing methods, detect disease and measure its ripeness. We will finally estimate their harvesting timing and let farmers know. All necessary devices will be provided.

This project includes to:

Project scope will be decided after the meeting with supervisors. You will bring to the role a passion for research and engineering, excellent computing skills (including a high level of programming ability), and a strong sense of responsibility. It will be better if you have an experience on vision processing and/or using DNN.

Location: Radio Systems 405.736

This project will develop a system to monitor the change of plant growing. 3D reconstructed model is required to estimate the harvesting timing as well as monitoring its growing. All necessary devices will be provided.

This project includes to:

Project scope will be decided after the meeting with supervisors. You will bring to the role a passion for research and engineering, excellent computing skills (including a high level of programming ability), and a strong sense of responsibility. It will be better if you have an experience on vision processing and/or using DNN.

Location: Robotics 405.652

This project will understand English sign-language from hand motion, gesture, and facial emotion. There are a lot of research on recognizing alphabets and numbers, but not word and sentence itself, which is more realistic and useful in real-world. Do literature survey on conventional approach, and select the best one and develop it as the first version. Then improve the method as the second version and/or integrate them to understand nuance of sign language. You may use reinforcement learning and/or Deep Neural Network (DNN). This project may be combined with other related projects, such as social human robot interaction projects and healthcare robot projects. All necessary devices will be provided.

This project includes to:

Project scope will be decided after the meeting with supervisors. You will bring to the role a passion for research and engineering, excellent computing skills (including a high level of programming ability), and a strong sense of responsibility. It will be better if you have an experience on vision processing and/or using DNN.

Location: Robotics 405.652

This project will develop a DNN model that synthesizes a sequence of face images that correspond to given speech semantics. Do a literature survey on conventional approaches and develop a face image generation system. All necessary devices will be provided.

This project includes to:

The project scope will be decided after the meeting with supervisors. You will bring to the role a passion for research and engineering, excellent computing skills (including a high level of programming ability), and a strong sense of responsibility. It will be better if you have experience with vision processing and/or using DNN.

Location: Robotics 405.652

This project will develop a multi-robot system using a guide robot and a logistics robot, which are already developed by part 4 students. We have three robots (a receptionist robot EveR, a guide robot Silbot, and a delivery robot GoCart), which can communicate with building system to control doors and elevators. This project will improve the integrated system focusing on navigation and/or interaction. You will work with other researchers who have done this project.

This project includes to:

Project scope will be decided after the meeting with supervisors. You will bring to the role a passion for research and engineering, excellent computing skills (including a high level of programming ability), and a strong sense of responsibility.

Location: Robotics 405.652

Wearable smart watches are increasingly being used for the detection of different health conditions by using machine-learning models to make inferences. We are, in particular, interested in the UoA developed MoodAI platform, which is developed for real-time mood state classification, based on the data obtained from Fitbit smart watches. The current platform, while being able to perform mood state classification using deep learning models, has some limitations. First, it only deals with data from Fitbit devices. Second, the ML models are black-box in nature and hence not explainable.

Wearable devices from many different vendors have their APIs and hence accessing wearable data from a range of wearable smart watches remains challenging. In this project, we seek to develop an unified interface that will enable data capture and processing from many different smart watches. More importantly, we will develop explainable deep ML models, which not only detect the onset of depression but also identify the key reasons for such a diagnosis, on an individualised basis. This project will be jointly supervised by Dr. Fred Sundram, who is a clinical psychiatrist and is based at FMHS.

Location: Embedded Systems 405.760

We have received funding from Google to collaborate on a new protocol for data centres called bittide. Bittide is a novel architecture for distributed systems, proposed by Google. Bitide is a physical layer ptotocol, which achieves "logical synchrony", without the need for actual clock synchronisation protocols. Logical synchrony ensures that buffers between communicating nodes are always bounded and also the communication delay is logically bounded.

In this project, we will develop an emulation platform of a network of bitide nodes. We have access to the hardware design from Google and need to create an example system comrising of multiple FPGAs communicating using bittide. We need to then implement startup protocol. Finally, we will work some some sample applications to execute over bittide.

References:

Lall S, Caşcaval C, Izzard M, Spalink T. Modeling and Control of bittide Synchronization. In2022 American Control Conference (ACC) 2022 Jun 8 (pp. 5185-5192). IEEE.

Roop PS, Pearce H, Monadjem K. Synchronous neural networks for cyber-physical systems. In2018 16th ACM/IEEE International Conference on Formal Methods and Models for System Design (MEMOCODE) 2018 Oct 15 (pp. 1-10). IEEE.

Location: Embedded Systems 405.760

Applications of Transformer-type deep learning models (such as those used in ChatGPT) continue to grow. They are well suited to process sequential or streams of data, like characters in a sentence. However, they can also capture patterns in non-language applications, like electrocardiograms (ECGs) from the human heart. ECGs capture the electrical signal of the heart and are used by clinicians to diagnose many heart conditions, however their analysis and understanding typically requires years of practice and knowledge.

This project will perform a design space exploration, and subsequent implementation, for a machine learning pipeline using Transformer-type neural networks to analyse and/or generate ECGs for a range of healthy and unhealthy hearts. The analysis aspect could improve the traditional signal-processing techniques which often fail to detect every feature of an ECG to reduce the burden on the clinician. Meanwhile, generated ECGs could be used to greatly increase a dataset being used to train and test ECG classifiers and pacemakers.

Location: Embedded Systems 405.760

Music can be beautiful, messy, and emotive, but it almost always tells a story. This story is created by a composer who writes in the language of music. Popular language models have begun to aid people in writing their own stories by taking in a few words to a sentence and predicting the next set of words. While this kind of technology is proficient in aiding people to write stories in English, a question emerges: Can we use language models to help composers write music?

This project will investigate developing a tool, potentially tailoring language models like GPT-3, to aid a composer in writing music by predict future notes based on the current notes. We envisage that the models will differ depending on music genre and so successful students will likely need to restrict themselves to a single genre that is traditionally well-structured, such as classical music.

Location: Embedded Systems 405.760

The software industry lacks diversity. Recent research has theorised that software development tools can have inbuilt biases, making them easier to use for people who think in a certain way. For example, GitHub, a commonly used software tool, has been found to have missing information, unclear terminology, and unexplained processes. This can make GitHub non-intuitive for people with process-oriented learning styles as they prefer to have complete information before starting a task.

This project will select a common software development tool or practice and investigate potential inbuilt biases in-depth by designing a lab experiment.

Location: HASEL 405.662

Introduction: With the increasing popularity of dating platforms such as Tinder or Bumble, it has become clear that these platforms play a significant role in shaping our dating and partner-finding experiences. However, the goals of these platforms do not always align with those of the users, and their impact on society has not been thoroughly researched.

Objective: Our goal is to develop a tool that will help us understand the behavior of dating platforms and the impact of their machine learning-based matching algorithms on society. We aim to do this in collaboration with researchers in interpersonal relationships at McMaster University.

Methods: To achieve this goal, we plan to carry out the following steps:

This will allow us to conduct experiments with volunteers and analyze the data gathered in these experimetns to gain insights into the impact of dating platforms on society. Conclusion: This project has the potential to provide valuable insights into the behavior of dating platforms and the impact of their algorithms on society. The data generated by our experiments will contribute to a better understanding of this important issue and pave the way for future research in this field.

Location: MDLS 405.559

mmWave is an emerging technology and has several benefits in comparison to existing sensors. However, it is currently rarely used for smart environment applications. This project aims to explore the capability of mmWave radar in terms of its range and precision for the purpose of performing indoor human activity recognition. The team will need to study and investigate the performance of mmWave radar against common human activities. The mmWave based results will also need to be compared against typical optical sensor (e.g. camera or lidar) based approach.

Location: Embedded Systems 405.760

Digital twin is an emerging technology that offers the ability to bridge the physical and cyber world, and provide intelligent services to its end users. This project aims to explore digital twin technology for its use in typical intelligent environment (smart home, smart office, smart building) applications. The project team will need to investigate and understanding state of the art digital twin technology, and design a digital twin platform for the intelligent room lab, targeting a smart office scenario. The digital twin platform will need to be able to provide continuous monitoring of the physical environment and providing intelligent/predictive control for its end users.

Location: Embedded Systems 405.760

Software Engineering has moved from the domain of single programmers working on a codebase to teams developing together. The recent lockdowns and the rise of modern web-based tools have also changed how developers work together. Developers can now collaborate in real-time on a codebase, even on individual files simultaneously. But what effect does this have on their productivity and the final code they produce?

In this project, you will explore how people use online, collaborative code editors. You will run evaluation studies where you observe how people work together and make recommendations on how to improve collaborative code editors.

Location: Radio Systems 405.736

Perceptual abilities can be affected by various conditions and also naturally decline with age. However, it has been shown that perceptual training can be used to mitigate decline and for rehabilitation.

AR and VR are increasingly being used to diagnose, treat and manage a wide range of health issues. However, a significant portion of these virtual diagnostic and treatment tools have been designed and developed for specific use cases. There exists little prior work that has incorporated a wide range of tools and techniques into a single platform to deliver rehabilitation services.

In previous work we developed in collaboration with researchers from Audiology bi-modal (audition and vision) AR and/or VR platform to deliver rehabilitation services to individuals diagnosed with tinnitus and balance disorders. In this research we will improve the prototype to make it more suitable for clinical research. This includes an interface for clinical users, improved interaction mechanisms within the virtual environment, different sound objects, and sound visualisations.

Location: MDLS 405.569

Wondering behaviour often occurs with senior adults who are suffering from dementia with a cognitive impairment. For example, Alzheimer’s disease erase a person’s memory so that familiar surroundings become unfamiliar. This disorientation often leads to wandering or lost, which is a common and serious concern for doctors and caregivers. Tracking devices have been introduced to locate the patients. These devices use either radio based frequencies or GPS signals to identify users' whereabouts. This project aims to develop a mobile based application that utilizes the data collected by wearable trackers and apply data analytics into the AI expert domain for elderly care. The indoor tracking system requies high accurracy/precision in terms of behavioural pattern matching, which often adopts radio frequency based mechanisum. Acitivities related information will be collected and analysed to determine patient risk factors. The study has a potential to help clinicians and researchers to have insights into rest home patients and their care management, so as to improve the overall healthcare practices.

Location: MDLS 405.569

Multi-modal learning uses multiple representations and interactions for supporting learning. For example, mathematics can be taught by using number representations, symbols (Latin numbers), or objects (both physical and virtual). It has been shown that many students learn better by using multiple representations and interactions (seeing, talking, doing). In this project we investigate whether tangible Augmented Reality (AR) can be used for learning data structures. Tangible AR combines real world objects with virtual augmentations. For example, students might use wooden blocks or coloured pieces as list elements and the program augments them with virtual information, e.g., links between elements. In this project we will develop a Tangible AR teaching tool for data structures and evaluate it with stage 1 students.

Location: MDLS 405.569

Spatial skills are a significant predictor of achievement in STEM subjects (Science, Technology, Engineering, and Mathematics). Furthermore, it has been shown that improvements in spatial reasoning skills, can increase achievements in mathematics. It is hence of interest to be able to test spatial skills to predict students’ performance, adjust learning content, and offer interventions to overcome any deficits in spatial intelligence.

While many spatial skills tests exist, they usually have only one version and hence can be learned and students can share answers.

In this project we will implement a web-based tool using procedurally-generated spatial skills tests, i.e., for each question type the tool can generate arbitrary many versions of it. We will then validate these tests with a user study by comparing results with a traditional spatial skills test.

Location: MDLS 405.559

A unmanned surface vehicle (USV) is powered by two jet pumps. Control of the direction and speed of the USV is established by adjusting the speed of each jet pump individually. Unfortunately the roll of the USV also needs to controlled and this can be achieved if the thrust vector can be determined.

The laboratory has purchased a commecial sensor but this project will develop and build another 3D force sensor using strain gauges (SG) attached to the transom and/or hull of the USV. The sensor will be calibrated and a model will be developed and compared using the commercial sensor.

This project will investigate the impact of ADC noise and precision on classical SG configurations. A maltese cross configuration is intended but other alternatives could also be investigated. A custom load cell for 3D force sensing will be constructed and models for transduction will be developed.

The project will require electronic and embedded programming and will suit EEE or EEE+CSE students.

Location: Control Systems 405.722

This project will develop, software and hardware, to operate a Gough-Stewart platform. The structural elements of platform are available. The project will integrate power amplifiers (PWM), linear actuators and embedded software to operate the platform in accordance with an arbitrary 3D profile.

You will start by operating each of the actuators in openloop. This will require you to write code for a PSoC and undertake measurements on each actuator. These measurements will be useful in building a Matlab simulation. The project will deliver a fully operational GS Platform that will be operated from a Matlab (or similar) GUI.

Location: Control Systems 405.722

Spatial skills are a significant predictor of achievement in STEM subjects (Science, Technology, Engineering, and Mathematics). Furthermore, it has been shown that improvements in spatial reasoning skills, can increase achievements in mathematics. In contrast to many other cognitive abilities, spatial skills can be trained.

While several spatial skill straining tools exist already, many of them are not engaging enough and/or do not cover all types of spatial skills (visual perception, spatial visualisation, mental rotation, visuospatial memory).

The goal of this project is to create an effective and engaging game for spatial skills training. Ideally the game should have a VR and non-VR interface so that we can compare the effectiveness of these two interfaces.

Several game prototypes exist and could be further developed for this research, or a new game could be created from scratch.

Location: MDLS 405.559

Aged care has become a popular domain for computelised applications. More and more sensor/wearable devices have been delopyed to rest homes as well as elderly care facilities in order to provide a smart enviroment of personalized care. Those smart systems not only dramatically reduce the physcial workload of the carefgivers but also offer an artificial intelligence driven precision health care. This project extends an existing implementation of a senior citizen tracking mobile application with additional functionalities, e.g., patient physiological data collectioin, behavourial monioring, sleep pattern analysis, etc. The aim is to develop a sofware solution with hardware support for smart home care. Acitivities related information will be collected and analysed to determine the patient risk factors. The study has a potential to help clinicians and researchers to have insights into rest home patients and their care management, so as to improve the overall healthcare practices.

Location: MDLS 405.569

Health and Safety (H&S) is a major concern in construction projects as the potential for death or injury is high and accidents occur at a higher frequency than other industries. One approach to help workers understand potential hazards is to provide visualisations of the work being undertaken on a site during the day. These visualisations typically show the position of safe zones for the site on a daily basis, based on the major work being undertaken. However, a construction site is very dynamic workplace with vehicle movements, crane load movements, temporary equipment being installed or dismantled (e.g., scaffolding) and the impact of these is not easily visualised in a 2D plan of a site. This project will look to draw together a range of important data sources to help visualise the real-time Health and Safety status of a construction site. This will include the 3D building model of what is currently built, the 4D schedule of material movements and construction for the site, sensor-based information on vehicle and crane movements, etc and visualise these in a VR, or AR, environment for the site health and safety officer.

Location: MDLS 405.569

IoT sensors are widely spread and collecting enormous amount of data. However, sensor data may contact noises caused by different factors, which will impact the data processing steps later on for providing intelligent services. This project aims to explore the existence and the type of anomalies (i.e. abnormal data) in IoT sensor time series, and investigate the potential to identify and classify them.

Location: Embedded Systems 405.760

ChatGPT is a conversational AI language model developed by OpenAI. It is based on transformer architecture and has been trained on a large corpus of text data from the internet to generate human-like responses to text-based prompts. It has demonstrated strong performance in generating coherent and natural-sounding responses to text inputs. Pepper is a humanoid robot and is designed to interact with humans and perform various tasks, such as answering questions, providing information, and performing actions based on voice commands. We would like to use the ChatGPT as the brain of a pepper robot to communicate with people.

This case study aims to observe human interaction with a robot. The focus is on capturing individuals' responses as they interact with a companion or stress management assistant robot (students and supervisors would choose the topic). The study will utilise the Pepper robot as the interface and gather responses from participants as they interact with a Chatbot, ChatGPT, or a human representative without the participants being aware of the spokesperson's identity. This project will be carried out in collaboration with the school of Psychology.

Location: Robotics 405.652

Music is a powerful motivator and is extensively used in gyms to make exercises more enjoyable and motivational. Music rhythm can help to increase physical activity by encouraging users to perform exercises, such as running, with the rhythm of the music. Several rhythm games, such as Dance Dance Revolution and Beat Sabre have become commercial successes.

Most existing rhythm games have been designed for home used. To encourage outdoor exercises mobile AR games have been developed. In this research we will investigate whether concepts from rhythm games can be combined with concepts of mobile AR games to increases users’ motivation, enjoyment, and performance when doing outdoor exercises.

We will develop a simple mobile AR rhythm game and investigate its effectiveness with a user study. The software should be able to measure users’ step frequency and the music rhythm. The game play should encourage increased synchronisation between step frequency and music rhythm using a scoring / reward mechanism and appropriate visual and auditory feedback.

Location: MDLS 405.569

With the recent weather events in Auckland, and many other recent disasters around the world, there has lacked a central online place where those affected can go to find out specific, up-to-date information (for example, a map of flooding-affected areas and road closures). This information exists online but is often in a variety of formats and in multiple locations.

In this project, you will investigate the use of modern data gathering and analysis techniques to automatically (as much as possible) curate and summarize important information related to disasters. The ideal end goal will be a prototype app or website where users can go to be presented with the most relevant, up-to-date information based on their location.

Location: MDLS 405.569

Headlines assert that “multitasking makes you stupid” but no causality has been demonstrated between multitasking and cognition. If multitasking with technology is changing the very nature of how we think—particularly for the worse—it is imperative to find out so that we can address negative effects. Even before COVID-19 prompted increases in technology use, US adults spent on average 11 hours per day consuming media but as few as 11 seconds focused on any given task before switching to another. Multitasking is becoming the norm but has been correlated with poor cognitive control (i.e., fundamental executive function including switching tasks and managing working memory).

Two major challenges with studying multitasking is 1) getting objective data on how much people multitask, and 2) changing multitasking behaviour in order to study the effects. In this project you will measure and parse objective multitasking data and help to conduct an intervention study to explore the cognitive effects of reduced multitasking.

Location: MDLS 405.569

This project is to iteratively refine a participative Unity-based installation where users experience immersive sound and visual feedback wearing a motion capture suit and VR headset. The immersive application directs gallery-visitors’ attention in various ways with sound and particle effects (for example: https://www.youtube.com/watch?v=wb_jM2kbqTA). The practical goal would be to refine the development of a complete gallery installation experience. The research goal would be to understand how the application affects sensory perception, embodiment, and subjective experiences. Over the year, we would customise the application based on stakeholder feedback.

Location: Power Systems 405.628

Worked examples have been shown to be one of the most effective ways of learning well-defined tasks. However, worked examples are time-consuming to produce, and many courses do not provide students with a wide range of worked examples to use in learning. This project aims to develop an online educational environment that will leverage a large language model (LLM) to automatically provide worked examples for standard programming tasks. The worked examples should use well-defined steps to illustrate the appropriate problem solving approach. The learning environment should include interactive components where possible. This project has the potential to have significant impact on student learning.

This project will be co-supervised by Dr Juho Leinonen, who is a postdoctoral fellow visiting the University of Auckland in 2023

Location: Control Systems 405.722

Large language models, such as ChatGPT and OpenAI Codex, have recently emerged as very powerful generative AI models. This project aims to integrate a large language model (LLM) directly within a pedagogic Integrated Development Environment (IDE) to support students learning programming. This project provides the option to extend an existing prototype tool, or develop a new tool from scratch. The existing prototype version of the IDE already has basic LLM support, and the goal of this project is to improve and refine the integration to provide more sophisticated and tailored suggestions and feedback to students. In addition to incorporating new features such as code completion, error detection and correction, and personalized feedback, the project will also aim to improve data collection in the IDE by collecting data at a more fine-grained level. By utilizing LLM support, students will be able to receive real-time feedback and guidance as they are programming, allowing them to write code more accurately and efficiently. The collected fine-grained data will provide valuable insights into students' coding behavior and learning, allowing for continuous improvement of the IDE and its LLM integration.

This project will be co-supervised by Dr Juho Leinonen, who is a postdoctoral fellow visiting the University of Auckland in 2023

Location: MDLS 405.559

Creating a large bank of multiple choice questions (MCQs) for student drill-and-practice, or for formal examinations, is difficult and time consuming. This project explores ways of using large language models (LLMs) to improve the efficiency of creating and maintaining these question banks. Starting with a set of learning outcomes, the software tool developed in this project should automatically create MCQs with sample solutions and explanations of why each answer is incorrect. Each aspect of the question (i.e., question stem, distractors, correct answer, explanations) should be editable by the instructor for quality assurance. The software tool should provide an efficient way to export and import such questions to existing systems (such as Canvas) using standard interchange formats, where they can be used for formal assessments, or to allow such questions to be answered by students for drill and practice. This project has the potential to have significant impact on teaching practice.

This project will be co-supervised by Dr Juho Leinonen, who is a postdoctoral fellow visiting the University of Auckland in 2023

Location: Control Systems 405.722

Machine Learning (ML) has become a crucial tool in solving complex problems across various domains and is highly valued by technology companies. As a result, there is a high demand for software engineers who have the skills to develop and implement ML models that can process and analyze vast amounts of data.

While the availability of ML libraries and frameworks has made it easier for software engineers to build ML models, it also requires a deep understanding of the underlying concepts and algorithms to ensure effective design and implementation. One of the major challenges in the field of ML is that its decisions and predictions are often not easily explainable. This is because many ML algorithms, such as deep learning neural networks, use complex processes to make predictions. This complexity makes it difficult for a novice to ML understand how the algorithm arrived at its conclusions and why it made certain decisions.

To address this challenge, explainability tools like LIME and SHAP have been developed to provide insights into the workings of ML models. This project aims to design and implement a computing education tool that leverages explainability to teach the key concepts of ML. The tool will be designed to work with any Python-based ML program and will rely on explainability tools such as LIME or SHAP to provide an intuitive and interactive way to learn about ML. By visualizing the workings of a ML model, students can gain a deeper understanding of how it makes predictions and solves problems, making the learning process more engaging and effective. The project will also experiment with the impact of bias or noise in data on the prediction.

This project is suitable for students who have a proficiency in the Python programming language or who are highly motivated to learn it.

Location: HASEL 405.662

There is an immediate need to develop technologies that can precisely capture the emergence and trajectory of peoples’ resilience in high-risk industries. Despite government efforts, high risk industries such as construction continue to lead work-related harm and fatalities. The integration of psychophysiological monitoring with virtual reality (VR) makes it possible to document an individual’s experiences as they interact in simulations of real-world contexts. However, conventional psychophysiological-VR systems are constrained to measuring limited features of behaviour in fixed VR environments. This project will develop a VR simulation of high-risk situations that can be experienced and engaged with by professionals working in the construction industry. A range of psychophysiological measures will be captured from the professional within the VR environment to help identify the most effective set of sensors and measures to identify a person’s risk resilience.

Location: MDLS 405.559

Machine Learning Testing is a recent research topic aiming at adapting software testing concepts to detect errors in ML-based software. One of the most challenging concepts to adapt is test oracle. In particular, it is challenging to automate the generation of a test oracle that can assess the quality of ML predictions beyond just passing all test cases. The oracle problem in machine learning refers to the challenge of defining a ground truth for evaluating the predictions made by a machine learning model. In traditional software testing, an oracle can be an expected output that is used to validate the correctness of a program's behavior. However, in machine learning, the prediction accuracy based solely on a set of labeled data may not be a reliable indicator of model performance on unseen data. This is because machine learning models often make predictions based on patterns learned from the training data, which may not generalize to new examples.

This research project aims to develop a novel automated method for defining "trustworthy" test oracles. Highly motivated students will explore how to use such oracles to automatically improve a ML model.

This project is suitable for students who have a proficiency in the Python programming language or who are highly motivated to learn it.

Location: HASEL 405.662

There is a need for more efficient evaluation of student-created artefacts, such as essays and programming assignments. Traditional methods of assessment, such as manual grading, can be time-consuming, subjective and hard to employ in large classes with thousands of students. Large language models (LLMs), such as ChatGPT and OpenAI Codex, have recently emerged as very powerful generative AI models. Using LLMs to assist in the evaluation of student-created artefacts may offer the potential to address this long-standing challenge. This project aims to develop a novel system that utilizes LLMs for assessing various types of student-created artefacts. The system will allow instructors to provide a rubric (e.g. marking guidelines) and example evaluations, and these will be provided to an LLM for the automatic evaluation of student-created artefacts. Students will be able to upload their own artefacts into the system and receive automated feedback from the LLM.

This project will be co-supervised by Dr Juho Leinonen, who is a postdoctoral fellow visiting the University of Auckland in 2023

Some relevant literature related to this project:

Automatic Generation of Programming Exercises and Code Explanations Using Large Language Models: https://dl.acm.org/doi/abs/10.1145/3501385.3543957

Location: MDLS 405.559

Software testability refers to the degree to which a software system or component can be tested efficiently and effectively. Software with a high-degree of testability is preferable as it is easier to write test cases, and thus easier to find software bugs. It is well understood that the effort of testing software systems depends on the artefacts under test, that is, some software systems are easier to test than others. Several studies have investigated the relationship between software metrics and test effort confirming that there exists a moderate correlation between software metrics and test effort.

The goal of this research project is to study how we could leverage such correlation to predict the test effort required to test a certain object oriented class. Early predicting

test effort can help developers to recognize refactoring opportunities to reduce the test effort. The project will investigate novel Machine Learning (ML) techniques to build predictive models for software testability.

This project is suitable for students who have a proficiency in the Python programming language or who are highly motivated to learn it. Indeed, Python is the de-facto programming language for building Machine Learning programs.

Location: HASEL 405.662

Program correctness can be automatically evaluated using test cases. However, the evaluation of programming style has typically required the input of an expert, such as a teacher. Recent advances in large language models (LLMs) such as GPT-3 used in ChatGPT have the potential to provide a level of automated feedback on programimng style that could be useful to both staff and students. This projects aims to investigate the effectiveness of automating feedback on programimng style through the analysis of program characteristics and LLM integration. This project has the potential to have significant impact on teaching of style in early undergraduate programming courses.

This project will be co-supervised by Dr Juho Leinonen, who is a postdoctoral fellow visiting the University of Auckland in 2023

Location: Control Systems 405.722

Many software systems rely on functionality developed by external libraries (dependencies). When libraries put out new releases, they could contain incompatibilities, called breaking changes, which could cause errors in the client projects if they change the dependency to use the new version. This is further complicated by the "chains" of dependencies that can exist. Dependencies can be direct dependencies or transitive dependencies. If a software project (the client project) uses a library's code directly, for example, through a method call, it is a direct dependency. If that method call then uses a method of another library, that is a transitive dependency. We do not know how much breaking changes in transitive dependencies impact client projects.

This project will analyse how and why breaking changes occur in a client project from transitive dependencies. For the analysis, you will use an existing dataset containing a set of client projects and their dependencies. The analysis will involve automatically updating dependency versions, examination of static analysis tool output, and analysis of source code.

Location: HASEL 405.662

The university is working to develop a Digital Twin (DT) of its buildings and facilities to support its sustainability goals. A digital twin comprises both a virtual model of a building and linkages to the performance of the physical entity, typically through sensor data. The virtual model can be used to simulate the impact of potential changes and then trigger those changes in the physical building. There are a significant number of data streams available about university buildings, covering data such as power use, water use, wifi connections, rubbish volumes, scheduled lectures, lighting, CO2 and temperature in spaces, airflow through HVAC systems, external weather station data, etc. While optimising the building use against each individual data stream is possible there is a need to identify how all the data streams interact and can be optimised in a holistic manner to meet the university’s sustainability goals.

Location: Green Electronics 405.712

Development Of Novice User Technologies: An AI-Assisted Tool for Learning Code Comprehension and Debugging

Code comprehension is an essential skill for students who are learning to program. With the recent emergence of powerful AI code generators (e.g. ChatGPT and GitHub Copilot), that can generate large amounts of output code such as entire functions and programs, the need for students to develop code comprehension skills is more important than ever. This project aims to build a tool that will help students to develop their code reading and comprehension abilities. The tool will automatically generate programs containing targeted sets of programming features appropriate for novice programmers. Similar to a standard debugger, it will allow students to step through the code illustrating the values stored in each variable. The novel aspect of the tool will be the integration of large language models that will be able to provide line-by-line explanations of the code as it is being executed. The goal is that this will provide direct educational benefits, as well as teach students how to make use of code debuggers and interpret their outputs. The tool will support dialogue-based interaction, using large language models, enabling students to ask for detailed explanations of steps they find confusing. Another pedagogical feature of the tool is that at various points, it will ask students questions to probe their understanding (e.g., what will the value of the variable be after the next line of code is executed?).

This project will be co-supervised by Dr Juho Leinonen, who is a postdoctoral fellow visiting the University of Auckland in 2023

Some relevant literature related to this project: Experiences from Using Code Explanations Generated by Large Language Models in a Web Software Development E-Book: https://arxiv.org/abs/2211.02265

Automated Questionnaires About Students’ JavaScript Programs: Towards Gauging Novice Programming Processes: https://dl.acm.org/doi/abs/10.1145/3576123.3576129

Automatic Generation of Programming Exercises and Code Explanations Using Large Language Models: https://dl.acm.org/doi/abs/10.1145/3501385.3543957

Location: Green Electronics 405.712

Our recent research has found a range of factors that software engineers consider when assessing the success of a software practice. Preliminary analysis indicates that these factors could be dependent on the personality of the individual software engineers. This project will further examine this relationship by developing and deploying an online questionnaire. Statistical analysis will be performed on the responses using the R statistical platform.

Location: HASEL 405.662

Massive amounts of information are produced when designing and constructing a building. Navigating this trove of information to identify the answer to a question about a specific part of the building and its design is a difficult task (e.g., what is the replacement life of this smoke stop door). New AI techniques support question/answering systems across a repository of information. This project will investigate the evolving AI techniques for question/answering and implement one for a particular use case of information querying about a building.

Location: Power Systems 405.628

The accurate measurement of an object's volume has important applications in industry and commerce. For example, freight charges are based on both weight and volume, but assessing the cost to the carrier can become complex for oddly shaped packages. Similarly, optimal packing of cargo holds requires an accurate description of object contours that go beyond assuming everything is a rectangular prism.

The purpose of this project is to develop a novel, low-cost, semi-convex volume measurement system that exploits basic ideas from computed tomorgraphy. The system should determine the volume in the time it takes for the object to pass through it.

The project is best suited to students interested in scanning and sampling systems, embedded systems (perhaps via FPGAs), and signal processing.

Location: Control Systems 405.722

The mathematical framework used to describe a continuous two-dimensional object in an inherently discretised machine vision system is called the "representation problem." Many possible solutions have been proposed over the years, each with advantages and disadvantages. One example is the so-called Fourier Descriptor representation where the 2D object is treated as a closed contour in the complex plane and modelled as a superposition of complex exponential basis functions. This representation has many attractive features but is best suited to very smooth object contours.

A recently discovered alternative representation, with distinct parallels to Fourier Descriptors, relies on complex polygonal basis functions (as opposed to smooth complex exponentials) that appears to be well-suited to objects with piecewise-linear contours.

This project will investigate the new object representation system and explore its features, properties, and applications as they pertain to machine vision applications.

This project is best suited to students with a strong interest in mathematical problem solving, signal processing, and image processing.

Location: Control Systems 405.722

Modern chess engines, such as Stockfish, are far stronger than even the best human players. Human players often use chess engines to analyse their games, find their mistakes, and identify stronger moves. However, the suggestions made by a strong engine are often not very human-interpretable, and therefore can be limited in their teaching utility.

The goal of this project will be to build an interface layer to a modern chess engine (i.e. Stockfish), to help the human user to interpret the recommended lines. The developed interface layer should identify common tactic and positional ideas in the recommended lines. It could also evaluate the complexity of a given line, or position, to help filter the engine recommendations for interpretability.

Location: HASEL 405.662

In the 2018/2019 financial year, there were 74,946 active claims with ACC for gym and fitness related injuries. The vast majority of gym-related injuries are self-inflicted as a result of poor technique when performing a given exercise, particularly with exercises such as squats and deadlifts. There are key metrics pertinent to performing these two exercises safely. Two of these metrics are force distribution between legs and relative knee angle throughout the exercise. We aim to implement a wearable sensor circuit to model a user’s legs during squats and deadlift exercises using surface electromyography (sEMG), mechanomyography (MMG), and near-infrared spectroscopy (NIRS), and comparing results against established methods and technology.

Location: Control Systems 405.722

Cryptography plays a vital role in securing many different types of information technology assets. Several encryption/decryption algorithms have been developed and employed in various applications to protect our sensitive data. This project investigates the security improvements achieved by dynamic reconfiguration on Field Programmable Gate Array (FPGA). A dynamic cryptosystem inspired by the principle of Dynamic Encryption [1] leads to creating different versions of encryption/decryption during the run. This approach will result in confusion for the malicious attackers and, hence, improved system security. This project investigates different methods of dynamic and partial reconfiguration of a cryptosystem on the FPGA platform and its impact on the system’s security.

[1] L. R. Knudsen, “Dynamic Encryption,” J. Cyber Secur. Mobil., vol. 3, no. 4, pp. 357–370, 2015, doi:https://doi.org/10.13052/jcsm2245-1439.341.

Location: Embedded Systems 405.760

Capsule networks (CapsNets) are a class of neural networks that address one of the main issues and deficiencies of CNNs. In CNNs, extracted features in initial layers pass through pooling layers where some important information, including spatial relationships, could be missed, resulting in incorrect classifications. CapsNets can better model hierarchical relationships and have shown a great performance in many fields. CapsNets have a different architecture in their layers and hence different computation complexity. This project investigates hardware accelerations for different layers of this class of neural networks for real-time applications.

Location: Embedded Systems 405.760

Recent advances in Machine Learning (ML) has resulted in widespread deployment of ML solutions in several areas. ML solutions are usually characterised by their high computational complexity. Heterogeneous computing platforms are introduced to optimise the performance and energy efficiency. They consist of a range of processing units, including Central Processing Unit (CPU), Graphics Processing Unit, Tensor Processing Unit (TPU), and Field Programmable Gate Array (FPGA). Each of the processing units on heterogenous platforms are sourced by its own specific voltage rail. This research will investigate the impact of undervolting on the accuracy of ML solutions implemented on AMD Xilinx heterogeneous platforms.

Location: Embedded Systems 405.760

Increasingly homes and businesses are investing in solar PV, high-energy battery storage systems and EV's. There is a need to develop mechanisms, tools and framework to purchase electricity not only from the energy retailers but also to trade peer-to-peer amongst consumers. Block chain architecture, using Ethereum hard-fork implementation of Smart Contract, was a pathway that was conceptualized, developed and trialled using transactive energy framework, through a 2017 Part IV power system project. The proof-of-concept was demonstrated, using the two UoA 3-kW grid connected solar system, battery storage, smart meter, PLC, micro-grid simulator, New Zealand wholesale real-time electricity market display etc. available at the Power systems Lab and interacting with UoA property services for grid services. A subsquent 2018 project developed a conceptual peer-to-peer that explored implementing the concept of time-value [as per the following publication Zhang, Z. J., & Nair, N.-K. (2017). Distributor “time value” pricing framework to schedule distributed energy resources. International Transactions on Electrical Energy Systems,.] This project explored integration of the developed concept using commercially available home-automation products (e.g. Apple Homekit, Nest etc.) through proprietary and standard-based protocols.

We expanded that understanding to expanded across a wide geographical area with visibility from a distribution utility with improved distribution management system (DMS) funcationality.

10.1109/peci48348.2020.9064650

Recently, we have also focussed on how customers can be integrated into this framework "Wavelet-Extreme Learning i-Machine for New Zealand Smart Meter Data" https://ieeexplore.ieee.org/document/9435262

The New Zealand Government has recently called for solutions to address the following objective that aligns with the previous projects that has been described earlier.

A 2021 project on peer-to-peer transactive platform for neighbourhoods was very successul and won a best paper award during 2022 EEA Conference. The part-4 project students closely worked with PhD student working in that area and helped gain recognition.

For 2023, we would like to embark upon developing a tool/platform so that we can enable participation by homes, neighbourhood and all electricity prosumers to participate in the NZ Electricity Reserve Market. SolarZero, who we support for research has successfully integrated a retail product of their fleet of solar-battery system to NZEM.

This project team will closely work with PhD and research group and the intent will be help roll-out a service or product by 2024 in New Zealand.

Location: Power Systems 405.628

This project is associated with exploring resilience impact assessment methods for regional electricity networks, using high-resolution high-impact weather scenarios, to help develop post impact recovery management and through some known case studies explore if any additional extra pre-preparatory information could have been extracted for future use.

One major source of electricity distribution network outage is due to wind-gust . As part of an ongoing research program and PhD projects, we have access to maximum wind gust analysis for Auckland, Waikato and Northland region across New Zealand on a granular. There is a need to focus attention on one geography like Auckland and develop co-metrics along with local distribution network company to find out the correlation of extended outages caused due to extreme wind gus scenarios of the past and then through some advanced learning and estimation techniquest like machine-learning, predict some potential outages based on future wind-data gust models.

Some work has already been done, which will be good starting reference for this project

(1) Resilience based Criticality Analysis for Seismic Performance Assessment of Underground Cables (2) Resilience Framework and Optimal Scheduling for DERs Factoring Uncertainties

(3) Extreme Weather Risk Framework for Power System Location-based Resilience Assessment

(Won Best Paper Award very recently)

The project team can also consider looking at other impact weather hazards like extreme flooding due to "atmospheric river" events that is affecting Auckland recently and accordingly arrange the framework.

Another potential case-study worth looking into could be forest fire. One of our recent interesting paper around forest fires and impact on solar generation is available from

Solar Power Generation in Riau-Indonesia, Potential and its Challenge Due to Forest Fire

Location: Power Systems 405.628

New Zealand is the “seabird capital of the world”, and these birds are a taonga. Seabirds are vital to the marine ecosystem, and many NZ species are threatened by human activities and climate change.

Tracking the movements of seabirds on the ocean is very challenging: birds can only safely carry 3% of their body weight; this limits the battery capacity, and rules out power-hungry GPS tracking for smaller birds.

Interestingly, flocks of seabirds are often “seen” on long-range marine radar systems, however the radar return from a single bird is very weak, and there is no way to identify individuals or a species.

This project will investigate the design of small, light-weight ‘reflectors’ that can be attached to a bird to enhance the strength of radar return.

Ideally the research team would fabricate and test their designs in a laboratory environment.

Location: Radio Systems 405.736

There are many forms of energy available in the ambient environment, such as radio frequency, solar, wind, thermal etc. However, only a small proportion of this energy is actually utilized to perform useful work. This leaves lots of spare energy distributed in very small quantities around us. This project proposes to investigate the levels of RF and solar energy that can be harvested from the ambient environement and then storing them in supercapacitors in an ultra efficient manner, to waste as little of the harvested energy as possible. The over arching aim to develop a self-sustainable IoT system which does not depend on battery power.

Location: Green Electronics 405.712

STEAM is a hot topic among parents and communities where there is a lot of focus on promoting Science, Technology, Engineering, Arts and Mathematics (STEAM) among the future generation. This project the aims to investigate the effectiveness of a series of activities allowing young children to explore the wonders of electronics and other hardware, giving them guided and well-informed exposure into engineering.

Location: Green Electronics 405.712

Modern Wireless power transfer magnetic designs are focused on both increasing power transfer while managing losses. Many new materials have been proposed which include soft magnetic composites made from crushed ferrite and polymer though to Nanocrystalline that may or may not have creaked surfaces to help reduce eddy currents. These represent materials with relatively low permeability though to very high permeability with varying conductivity.

When placed into a wireless pad, they necessarily have to be shaped into tiles that have some gaps between them. Their respective capability to manage high flux density or eddy currents mean that their position in an inductive power transfer (IPT) pad, has to be carefully selected, as some may be better for managing leakage fields while others are suitable for directing the main field that couples the ground side and vehicle magnetics to help transfer power.

Most often these materials are tested using toroid’s with unidirectional fields, but IPT fields can enter and exit the magnetic structure of the pad in all directions and the materials can have varying flux densities. Thus assessing the best position and use of these materials in proximity to the coil is critical.

Gapping magnetic cores into tiles also helps manage magnetic flux density and eddy current and helps avoid core saturation.

This project seeks to build on previous work that began to assess the effects of the Litz wire coil in close proximity to these gapped magnetic cores in an IPT magnetic pad and to help identify best placement, and gap relative to the power coils and suitable regions where the available SMC and nanocrystalline structures should be used. Ansys can be used to broadly understand field levels but loss will need to be evaluated experimentally.

Location: Power Electronics 405.614

Web3 and blockchain technology are growing in popularity and have interesting applications to offer for web development. The smart contract is the core of Web3 that allows us to interact with the blockchain. To enable interaction, a smart contract must be developed, tested and deployed on a network. Therefore, having a framework that is easy to use is extremely important. Some existing frameworks (like Hardhat) already have a comprehensive toolkit and are used by the majority of developers. However, they have potential downsides: for example, Hardhat only allows developers to use scripts to deploy and test smart contracts on their local chain. It would be more user friendly if a user interface like Etherscan allowed developers to interact with the contracts. However, Etherscan is only available on testnet and mainnet.

In our project, we will examine and analyze existing frameworks from the perspective of smart contract developers and potentially discover further usability problems. We will also interview smart contract engineers in the industry to understand Web3 development further. Throughout this project, we aim to develop and deploy our own blockchain development framework that will solve several existing issues.

Location: MDLS 405.559

Playing a percussive instrument such as drumming requires the musician to recall their practice from vision, sound and sense of touch. Remote teaching of finger drumming in virtual reality currently relies largely on vision and sound. It may benefit from the inclusion of shared haptic feedback between teacher and student. We would like to investigate the effects of incorporating haptic feedback on engagement and empathy between the teacher and student.

Human Computer Interaction has been an area of research for several decades. More recently, there has been a growing amount of research on the effects of shared haptics in VR on empathy. Work that has been done includes research on the complementary effects of haptics and audio [1], on the perception of rhythm[2]. It has also investigated whether the ability to detect changes in haptics change has the user become familiar at discriminating between different textures, creating a haptic “muscle memory”. This could be potentially valuable for more effective remote teaching of a rhythm based task.

Both the teacher will have small haptic devices attached to each of their fingers, and as the teacher taps their fingers, the student will feel vibrations on the corresponding fingers.

We will measure the effectiveness of teaching in two conditions:

Condition 1: Audio + Haptics

Condition 2: Visuals + Audio + Haptics

The effectiveness of teaching will be measured based on accuracy of the rhythm, i.e how close in timing, and this will be represented in the form of a score for comparison.

Location: Power Systems 405.628

Machine Learning (ML) has become a crucial tool in solving complex problems across various domains and is highly valued by technology companies. As a result, there is a high demand for software engineers who have the skills to develop and implement ML models that can process and analyze vast amounts of data.

While the availability of ML libraries and frameworks has made it easier for software engineers to build ML models, it also requires a deep understanding of the underlying concepts and algorithms to ensure effective design and implementation. One of the major challenges in the field of ML is that its decisions and predictions are often not easily explainable. This is because many ML algorithms, such as deep learning neural networks, use complex processes to make predictions. This complexity makes it difficult for a novice to ML understand how the algorithm arrived at its conclusions and why it made certain decisions.

To address this challenge, we will explore how to use ChatGPT and similar tool to help novice ML programmers to learn ML concepts.

This project is suitable for students who have a proficiency in the Python programming language or who are highly motivated to learn it.

Location: HASEL 405.662

Machine learning has gained a lot of popularity over the years. This project will involve identification of an educational domain that will benefit from applying machine learning techniques. After the literature review is carried out, you will design, build, and test a software solution that employs machine learning for an educational domain to be decided

Location: Control Systems 405.722