Computing and Mathematical Sciences Papers

Permanent URI for this collectionhttps://researchcommons.waikato.ac.nz/handle/10289/6

This collection houses research from the School of Computing and Mathematical Sciences at the University of Waikato.

Browse

Now showing 1 - 20 of 1568

Simple lambda lifting: Formalisation in Lean and a new efficient algorithm
(ACM, 2026) Levy, Tom; Reeves, Steve
Lambda lifting is a technique used in compilers to convert nested function definitions to top-level function definitions. A series of papers has led to an 𝑂(𝑛2) algorithm, however it is complex. We present a simple 𝑂(𝑛2) algorithm for lambda lifting and prove its correctness. We also formalise a lambda lifting specification from the literature in Lean 4, and use that to prove some of the properties and test our algorithm on generated test cases. One of our contributions is to formalise the notion of a “complete” and “minimal” lifting, addressing a small issue with the handling of unused functions that to our knowledge affects all previous algorithms.
Improving glycaemic control in primary care for Tongan adults with type 2 diabetes through the use of continuous glucose monitoring and holistic support: a pilot study
(CSIRO Publishing, 2026-04-27) Galewski-Tangataevaha, Janina’ofa; Crocket, Hamish; Aporosa, S Apo’; Vaka, Sione; Yoon, Seong Hoon; Chepulis, Lynne; Stokes, Tim
Introduction In Aotearoa New Zealand, Pacific peoples, including Tongans, experience disproportionately higher rates of type 2 diabetes and related complications. There is an urgent need for innovative, culturally appropriate interventions to improve outcomes. Aim This study aimed to determine the impact of continuous glucose monitoring devices with cultural wrap-around support on medium-term glycaemic control and other type 2 diabetes biomarkers in Tongan adults with high-risk type 2 diabetes. Methods Twenty-two Tongan adults with HbA1c 60 mmol/mol were invited to participate in a 6-month pilot intervention study involving 4 weeks of continuous glucose monitoring wear at baseline and 2 weeks at 3-months, alongside wrap-around care delivered by a Tongan kaiāwhina (support health worker). The primary endpoint was 3-month HbA1c. Clinical (glycated haemoglobin, lipids, estimated glomerular filtration rate, urinary albumin to creatinine ratio) and psychosocial (Diabetes Self-Management Questionnaire, measured at baseline and 3 months) outcomes were measured at baseline, 3, and 6 months. Results Nineteen participants completed the study through to 6 months. Mean HbA1c significantly decreased from 80.2 ± 19.4 mmol/mol at baseline to 68.6 ± 14.2 mmol/mol at 3 months, with reductions maintained at 6 months. No significant changes in lipids or renal function were observed. Diabetes Self-Management Questionnaire scores increased from 4.9 ± 0.8 to 6.0 ± 1.0 (P < 0.001). Discussion Culturally tailored continuous glucose monitoring-based interventions have the potential to support Tongan adults with understanding, optimising, and managing type 2 diabetes.
Repeated augmented rehearsal: A simple but strong baseline for online continual learning
(NeurIPS, 2022) Zhang, Yaqian; Pfahringer, Bernhard; Frank, Eibe; Bifet, Albert; Lim, Nick Jin Sean; Jia, Yunzhe; Koyejo, S; Mohamed, S; Agarwal, A; Belgrave, D; Cho, K; Oh, A
Online continual learning (OCL) aims to train neural networks incrementally from a non-stationary data stream with a single pass through data. Rehearsal-based methods attempt to approximate the observed input distributions over time with a small memory and revisit them later to avoid forgetting. Despite their strong empirical performance, rehearsal methods still suffer from a poor approximation of past data's loss landscape with memory samples. This paper revisits the rehearsal dynamics in online settings. We provide theoretical insights on the inherent memory overfitting risk from the viewpoint of biased and dynamic empirical risk minimization, and examine the merits and limits of repeated rehearsal. Inspired by our analysis, a simple and intuitive baseline, repeated augmented rehearsal (RAR), is designed to address the underfitting-overfitting dilemma of online rehearsal. Surprisingly, across four rather different OCL benchmarks, this simple baseline outperforms vanilla rehearsal by 9%-17% and also significantly improves the state-of-the-art rehearsal-based methods MIR, ASER, and SCR. We also demonstrate that RAR successfully achieves an accurate approximation of the loss landscape of past data and high-loss ridge aversion in its learning trajectory. Extensive ablation studies are conducted to study the interplay between repeated and augmented rehearsal, and reinforcement learning (RL) is applied to dynamically adjust the hyperparameters of RAR to balance the stability-plasticity trade-off online. Code is available at https://github.com/YaqianZhang/RepeatedAugmentedRehearsal.
XGBoost: Scalable GPU accelerated learning
(2018) Mitchell, Rory; Adinets, Andrey; Rao, Thejaswi; Frank, Eibe
We describe the multi-GPU gradient boosting algorithm implemented in the XGBoost library (https://github.com/dmlc/xgboost). Our algorithm allows fast, scalable training on multi-GPU systems with all of the features of the XGBoost library. We employ data compression techniques to minimise the usage of scarce GPU memory while still allowing highly efficient implementation. Using our algorithm we show that it is possible to process 115 million training instances in under three minutes on a publicly available cloud computing instance. The algorithm is implemented using end-to-end GPU parallelism, with prediction, gradient calculation, feature quantisation, decision tree construction and evaluation phases all computed on device.
CodeWars: Using LLMs for vulnerability analysis in cybersecurity education
(The Colloquium for Information Systems Security Education, 2026-03-21) Chaudhary, Arunima; Colombo, Gualtiero; Javed, Amir; Haseeb, Junaid; Kumar, Vimal; Larsen, Richard
Large Language Models (LLMs) are increasingly explored as tools for software development and could further constitute a supplementary source for the development of varied examples intended for pedagogical use. While they can improve productivity, their ability to produce code that is both secure and compliant with Secure Software Development (SSD) practices remains uncertain, raising concerns about their role in cybersecurity education. If LLMs are to be integrated effectively, students must be trained to critically evaluate generated code for correctness and vulnerabilities, raising an important question: How can LLM-generated code be effectively and securely incorporated into Cybersecurity education for teaching vulnerability analysis? This paper introduces CodeWars, a novel teaching methodology that combines LLM-generated and human-written code to examine how students engage with vulnerability detection tasks. CodeWars was implemented as a pilot study with a total of 32 students at Cardiff University and the University of Waikato, where students analyzed flawed, secure, and mixed-origin code samples. By comparing student approaches, analysis, and perceptions, the study provides insights into how vulnerabilities are detected, how code origins are distinguished, and how SSD practices are applied. Our analysis of student feedback and interviews indicates that Codewars produced structured and accessible code, simplifying vulnerability identification and offering educators the means to efficiently develop varied SSD teaching applications. These findings illuminate both the advantages and constraints of employing LLMs in secure coding and position this study as a foundational step toward the responsible adoption of AI in Cybersecurity Education.
Empowering education: Student perceptions and attitudes to the role of social robots in learning contexts
(ACM, 2026-03-16) Turner, Jessica Dawn; Vanderschantz, Nicholas; Bowen, Judy; König, Jemma Lynette; Carino, Hannah
Successful integration of social robots in education relies on the acceptance of robots in learning contexts by students. Using a participatory design workshop, students interacted with a KettyBot and ideated potential roles for robots in the classroom. This was followed by a questionnaire and the Godspeed Questionnaire Series (GQS) to understand student perceptions and attitudes towards social robots in education environments. Learners described potential use cases and our results demonstrate students envision robots as assistants rather than teachers, emphasising the importance of human connection in learning.
“It almost wanted to hurt someone”: The impact of intentional creepiness on user perceptions
(ACM, 2026-03-16) Turner, Jessica Dawn; Vanderschantz, Nicholas; König, Jemma Lynette; Siddika, Rafeea
The intentional design of robots to evoke creepiness provides a unique lens for studying human perception and willingness to engage. To understand user perceptions and acceptance of robots we developed a robot prototype designed with targeted facial, morphological, and movement features that may be perceived as "creepy". Using the Human-Robot Interaction Evaluation Scale (HRIES) we found that disturbance was moderate towards our intentionally creepy robot with significant participant variation. Furthermore, qualitative results confirmed this polarity, with descriptions ranging from "angry and unfriendly" to "cool and cute". This variability demonstrates that "creepiness" is more subjective than initially anticipated and highlights a key research gap in academic literature with the need for measurement tools which capture negative perceptions in HRI.
Accuracy of machine learning models versus "hand crafted" expert systems – A credit scoring case study
(Elsevier, 2009) Ben-David, Arie; Frank, Eibe
Relatively few publications compare machine learning models with expert systems when applied to the same problem domain. Most publications emphasize those cases where the former beat the latter. Is it a realistic picture of the state of the art? Some other findings are presented here. The accuracy of a real world “mind crafted” credit scoring expert system is compared with dozens of machine learning models. The results show that while some machine learning models can surpass the expert system’s accuracy with statistical significance, most models do not. More interestingly, this happened only when the problem was treated as regression. In contrast, no machine learning model showed any statistically significant advantage over the expert system’s accuracy when the same problem was treated as classification. Since the true nature of the class data was ordinal, the latter is the more appropriate setting. It is also shown that the answer to the question is highly dependent on the meter that is being used to define accuracy.
Gene selection from microarray data for cancer classification - A machine learning approach
(Elsevier, 2005) Wang, Yu; Tetko, Igor V.; Hall, Mark A.; Frank, Eibe; Facius, Axel; Mayer, Klaus F.X.; Mewes, Hans W.
A DNA microarray can track the expression levels of thousands of genes simultaneously. Previous research has demonstrated that this technology can be useful in the classification of cancers. Cancer microarray data normally contains a small number of samples which have a large number of gene expression levels as features. To select relevant genes involved in different types of cancer remains a challenge. In order to extract useful gene information from cancer microarray data and reduce dimensionality, feature selection algorithms were systematically investigated in this study. Using a correlation-based feature selector combined with machine learning algorithms such as decision trees, nave Bayes and support vector machines, we show that classification performance at least as good as published results can be obtained on acute leukemia and diffuse large B-cell lymphoma microarray data sets. We also demonstrate that a combined use of different classification and feature selection approaches makes it possible to select relevant genes with high confidence. This is also the first paper which discusses both computational and biological evidence for the involvement of zyxin in leukaemogenesis.
10-year survival comparison of two cemented implants in primary total hip arthroplasty for osteoarthritis: A New Zealand regional study
(Springer, 2025) Pearce, Amy; Joshi, Chaitanya; Chan, Georgina; Lamberton, Tony; MacLean, Simon; Vane, Andrew; Hébert-Losier, Kim
Introduction Compare 10-year survival of the cemented highly crosslinked polyethylene Exeter® Rimfit™ (Rimfit) Cup and its predecessor, the ultra-high molecular weight polyethylene Exeter® Contemporary Flanged Cup™ (ECF), both with an Exeter® V40™ stem, in primary total hip arthroplasty (THA) for osteoarthritis in the Bay of Plenty region of NZ. Method We extracted national registry data for THA surgeries in the region between 1 January 2003 and 30 June 2023 and report the 10-year survival and reasons for revision of the two fully cemented implants (n = 495). We compared standard Kaplan-Meier estimates using the log-rank test. Cox proportional hazard models investigated the potential influence of six patient variables on the survival of each implant: sex, age, body mass index (BMI), ethnicity, American Society of Anesthesiologists (ASA) rating, and funding source (public/private). Results No statistically significant difference in 10-year survival rate between the implants (p = 0.334) (ECF 95.6% [93.4, 97.9], Rimfit 97.0% [95.9, 98.2]) or statistically significant difference in revision reasons between the implants (p = 0.09) was noted. Cox regression revealed no statistically significant influence of any of the six patient variables on the 10-year survival of the ECF (p = 0.584) or Rimfit (p = 0.611). Conclusion Both implants exceeded 95% survival at 10-years, which is favourable compared to the corresponding 94.8% national survivorship of cemented implants in NZ. There is no statistically significant difference in the 10-year survival rate or reasons for revision of the two cemented implants compared in this region. The Rimfit appears a suitable alternative to the ECF, from a survival and revision perspective.
15-year patient-reported outcomes of a cemented flanged cup and stem combination in primary total hip arthroplasty: A New Zealand study
(SAGE, 2026) Pearce, Amy; Joshi, Chaitanya; Chan, Georgina; Lamberton, Tony; MacLean, Simon; Vane, Andrew; Hébert-Losier, Kim
Methods: We investigated 15-year patient-reported outcomes (PROMs) and their predictors in primary total hip arthroplasty (THA) for osteoarthritis using a cemented flanged cup and stem from a regional joint registry in New Zealand. Regional data were collected for all primary THAs with this cemented combination from 1 January 2003 to 30 June 2023 who had recorded PROMs on at least 1 occasion (n = 263). PROMs included Oxford Hip Score, Western Ontario and McMaster Universities Arthritis Index and Veterans Rand-12, evaluated against patient age, ethnicity, sex, body mass index (BMI), funding pathway, and American Society of Anesthesiologists (ASA) rating. Results: Significant improvements across preoperative PROMs were noted 1-year post-surgery, with a mean change above 23 in the Oxford Hip Score maintained at 5, 10, and 15 years (p ⩽ 0.001). Conclusions: Regression analysis indicated that being female, public funding, and higher BMI were associated with worse preoperative PROMs. Poorer preoperative scores, older age and ASA 3 rating correlated with poorer postoperative outcomes.
Computer graphic art of clothing
(2019) Soo, Chin-En Keith
We are living in an ever-changing world, where new methods are being introduced to carry out the most staightforward task, where new inventions are being proposed to ease our daily operations, where new ideas are popping out at every corner. Since the dawn of time, mankind has been using innovation and creativity to survive and enhance life. With the use of technology, it has enabled more possibility and more significant endeavour. Now and forever, we are dependant on technology, that has played a substantial role in our design solutions, which inescapably affect every one of us.
Type brighter
(Domus Argenia, 2013) Soo, Chin-En Keith; Soddu, C; Colabella, E
Type Brighter is intended as a new way of reading the alphabet. Shape, colour and pattern create memorable sequences based on characteristics of the letterform. By utilizing colour and repetition, readability is promoted. Each letter of the English alphabet is assigned a colour and positions, resulting in a full set of unique patterns. The user types using the keyboard, and the corresponding lights are shown on the light board. The simplicity of colour makes Type Brighter an alternative to more complicated communications such as mores code, and the use of pattern creates memorable sequences of colour. User can also experience the change of ambience, while the moving colour type projects an abstract story using light. Type Brighter aims to create a new visual language through light. Colour, shape and pattern are strong visual elements, and when combined create a memorable experience. Colour serves an important part of our everyday lives and is easily distinguishable in all situations, making Type Brighter effective in a range of applications.
RenderRing
(Domus Argenia, 2015) Soo, Chin-En Keith; Soddu, Celestino; Colabella, Enrica
Renderring is a musical platform for intuitive composition. It enables users’ interaction to provide opportunity for anyone to draw a unique circle and translates the drawing into a piece of melody. Users are able to set the composition variables before they start (Tempo, time signature and number of notes). The process involves two parts: First is a collection of user input by getting user to draw any unique circle in a provided space. Second is an interpretation using the program to decipher the drawing and identify point of intersections on the musical staff. After which, the program will produce a unique piece of melody with the user’s drawing. The user can then proceed with options of redoing or saving the melody. Renderring aims to bring new experience to create melody with a vision to simplify complexity. Transferring oneself energy from one form to another by converting visual to sound. The process enables creativity and empowers everyone to express his or her hidden inner potentials by making straightforward music.
Hueue
(Domus Argenia, 2017) Soo, Chin-En Keith; Soddu, Celestino; Colabella, Enrica
HUEUE aims to capture the colour story of a movie and present it in an accessible time frame of a minute. Movies at their simplest are colour, sound and motion. HUEUE aims to distil any movie into these basic forms and generate a unique form of escapism, bring the audience on a journey into the movie itself. HUEUE creates a tunnel effect. The effect indicates an impression of a portal. This is intended to give life to the escapism and create a more concrete feeling of the journey with the aid of sequential colours flowing from the movie. The audio is condensed creating a pitch shift, simulating the Doppler effect. All these elements create an experience that accelerate the viewer into the escapism and further into the movie.
Palate 字觉
(Domus Argenia, 2018) Soo, Chin-En Keith; Soddu, Celestino; Colabella, Enrica
Chinese characters are a visual symbol with strong contagion. Palate uses the pronunciation and character structure of Chinese characters as the entry point to colorize the Chinese characters so as to give them the possibility of expressing colors in the design of Chinese characters. Each Chinese character has its own unique color system. The application of the colorisation of Chinese characters can help to study the artistic charm of Chinese characters from a new perspective, improve the visual impact of Chinese characters, break the limitations of the past in the search for changes in the design of Chinese characters, and seeking a new form of modern Chinese character design.
ebBe
(Domus Argenia, 2021) Soo, Chin-En Keith; Simmons, Rowan; Soddu, Celestino; Colabella, Enrica
ebBe is a visualisation of tweets in realtime. ebBe uses word frequency to express the notion and motion of an ebb by contracting lines representing creation and decay. The created lines in different qualities will mimic and manifest a live visual artwork.
Chromatic ghost
(Domus Argenia, 2024) Soo, Chin-En Keith; Simmons, Rowan; Soddu, Celestino; Colabella, Enrica
"Chromatic Ghost" delves into the convergence of artificial intelligence and the emotional narrative of cinema, exploring how palettes can AI-generated colour visually express the emotional depth of films. The project aims to assess the capabilities of AI in understanding and translating human emotions into visual representations, questioning how effectively machines, which are increasingly embedded in our "Chromatic Ghost" is a creative experiment designed to investigate how AI can interpret and visualise human emotion in the context of film. In an era where artificial intelligence is increasingly intertwined with our daily lives, this work examines how well these technologies can grasp and reflect the complexities of human emotions, which are often considered too nuanced or abstract for machines to comprehend. lives, can interpret the intricacies of our emotional experiences. By using AI to generate emotion- based colour palettes and applying them to film frames, "Chromatic Ghost" transforms iconic cinematic scenes into layered, ethereal images that evoke the emotional core of the films in a spectral, dreamlike manner. The resulting visual compositions—ghostly, fragmented, and nuanced—invite viewers to contemplate both films' emotional resonance and AI’s role as a mediator of this experience.
Co-WARe
(Domus Argenia, 2022) Soo, Chin-En Keith; Soddu, Celestino; Colabella, Enrica
Throughout history, humans have been creating receptacles in their daily activities to hold, keep, and preserve the rewards and objects they treasure. Applying the same notion, Co-WARe makes unique receptacles from covid data to express the information in an artistic form. Co-WARe is an objectified presentation of all COVID-19 cases, deaths, geographically located data in each country, and the time of the data was generated. These data series project the different changes brought to each country since the beginning of COVID-19. It also provides more intuitive insight into the epidemic in all countries worldwide.
A natural behavior planner for multi-personal human-robot interaction within the simulated environment
(Elsevier BV, 2026-03) Chen, Yue; Zheng, Pai; Zhou, Zhiyuan; Soo, Chin-En Keith; Wang, Haining; Yu, Chunyang
In recent years, diffusion models have made remarkable success in generating realistic human motions. However, existing robot pose-learning approaches are largely focused on single-task and one-to-one scenarios, failing to account for multi-person social interactions. This limitation leads to rigid, context-insensitive behaviors that are ill-suited for real-world service scenarios. Consequently, current systems often produce robotic behaviors incapable of the fluidity and responsiveness expected in human-centered environments, a shortcoming underscored by affordance theory in robotics. To address this issue, we propose RoboActor, an innovative human-robot interaction behavior planner that draws inspiration from theatrical acting to orchestrate both deliberate and automatic actions. Our framework leverages large language models (LLMs) to disentangle primary command-driven tasks from secondary, context-induced subtasks. By this means, RoboActor generates lifelike and socially appropriate behaviors in multi-person settings, significantly enhancing the naturalness, engagement, and realism of service robots in everyday social applications.

Browse

Recent Submissions