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 1666

Māori algorithmic sovereignty: idea, principles, and use
(Ubiquity Press, 2024-04-02) Brown, Paul T.; Wilson, Daniel; West, Kiri; Escott, Kirita-Rose; Basabas, Kiya; Ritchie, Ben; Lucas, Danielle; Taia, Ivy; Kusabs, Natalie; Keegan, Te Taka Adrian Gregory
With the emergence of data technologies and algorithms in Aotearoa New Zealand that are used for decision-making and support, there is a need for frameworks to guide how we maximise the opportunities these technologies create and minimise the risks they may impose. For algorithms that use Māori data, these require extra considerations due to the heightened risks Māori endure due to systemic biases inherent within data and the processes that underlie algorithm development. Algorithms can be framed as a particular use of data, therefore data frameworks that currently exist can be extended to include algorithms. Māori data sovereignty principles are well-known and are used by researchers and government agencies to guide the culturally appropriate use of Māori data. Extending these principles to fit the context of algorithms, and re-working the underlying sub-principles to address issues related to responsible algorithms from a Māori perspective leads to the Māori algorithmic sovereignty principles. We define this idea, present the updated principles and sub-principles, and highlight a strategy for the detection and minimisation of bias within the algorithm development process.
A heuristics matrix for evaluating the learnability of safety critical medical systems
(Elsevier, 2025) Turner, Jessica; Bowen, Judy; Moxham-Bettridge, Veela
The increasing use of medical devices in non-clinical settings has elevated the importance of learnability, a key aspect of usability, that is defined as how easily users can learn to operate a system. While learnability is recognised in international usability standards, its varied definitions and inconsistent application emphasise the need for more targeted evaluation methods. This paper addresses this gap by proposing a framework for assessing learnability based on user experience levels and context of use. We define a set of user categories and introduce a learnability matrix to evaluate how different types of users, such as patients, caregivers, and medical professionals, interact with medical technologies. To illustrate the methodology, we perform a case study using the Medtronic MiniMed 770G insulin pump system. Our analysis shows that while primary and secondary users are often well-supported, passive, infrequent or adhoc users may face challenges in learning to use the device effectively in emergency settings. These findings underscore the need for more inclusive design and evaluation approaches that account for multiple user roles, to improve the safety and usability of home-based medical technologies.
Robot student ambassadors: Using service robots for community engaged events
(ACM, 2026) Turner, Jessica
Universities are increasingly exploring how service robots can be used as tools for public engagement and outreach with communities. This work investigates the use of service robots deployed at outreach events such as open days and community events like STEM Festivals to promote the university and engage positively with the public. Two service robots, Cruz and Marshmallow, have been deployed across multiple community-based events. Observations focus on visitor interactions and explore the different uses for these robots in community settings. Early results demonstrate strong user engagement and facilitated discussion about university research. The presence of these robots lowered barriers for community members, particularly children, who engaged naturally through touch, speech and playfulness. Findings suggest that robot ambassadors can enhance community outreach and enrich campus experiences. They appear to be valuable for engaging with young audiences and supporting informal learning, generating awareness of cultural and community perspectives.
The impact of data aggregation: Advocating for individualized analysis in wearable sensor research
(ACM, 2025) König, Jemma; Penaredondo, Jascha; Lim, Nick Jin Sean; Hinze, Annika; Bowen, Judy; Viller, Stephen; Paay, Jeni; Fredericks, Joel; Turner, Jane; Vickery, Nicole; Wadley, Greg; Muñoz, Diego; Capel, Tara; Atiq, Arzoo; Davis, Hilary; Boden, Marie; Harman, Joel; Ploderer, Bernd
The use of wearable sensors to record physiological data is becoming increasingly common, both in research and in daily life. User studies collect data using wrist bands, chest straps, and headbands to measure heart rate, skin conductance, and brain activity, to name a few. These readings can then be used to classify a range of physical and cognitive functions, such as cognitive and physical workload, cognitive and physical fatigue, stress, and attention. However, while physiological data is highly individualized, many researchers use machine learning classification methods that combine the participants' data into one dataset. In this paper, we demonstrate the negative impact this has on results and show that the individualized nature of physiological data requires individualized analysis and classification.
Categorical representation of DRC-semigroups
(Elsevier, 2026) East, James; Fresacher, Matthias; Azeef Muhammed, P. A.; Stokes, Tim E.
DRC-semigroups model associative systems with domain and range operations, and contain many important classes, such as inverse, restriction, Ehresmann, regular ⁎-, and ⁎-regular semigroups; concrete examples include diagram monoids, linear monoids, relation monoids, among many others. In this paper we show that the category of DRC-semigroups is isomorphic to a category of certain biordered categories whose object sets are projection algebras in the sense of Jones. This extends the recent groupoid approach to regular ⁎-semigroups of the first and third authors. We also establish the existence of free DRC-semigroups by constructing a left adjoint to the forgetful functor into the category of projection algebras.
Exploring opportunities for flexible wearables to support physical training
(ACM, 2025) Stawarz, Katarzyna; Everitt, Aluna; Bowen, Judy
Advances in digital fabrication with flexible and conductive materials are enabling new opportunities for customisable wearables that enhance physical training across various sports through on-body posture and movement monitoring. However, before developing new wearable devices, it is crucial to understand the unique challenges and opportunities of real-world use and the diverse needs of stakeholders such as athletes, trainers, and sports enthusiasts. In this paper, we report results of two co-design workshops conducted with 11 participants representing a range of sports backgrounds and interests to identify challenges and opportunities for flexible wearables. The participants’ designs and prototypes highlight the importance of supporting personalised wearable design, addressing varying contexts, and accommodating different skill levels. We propose design considerations for wearable devices that prioritise flexibility, both in their materiality to support a wide range of movement and in their adaptability to accommodate changing conditions and user progression across skill levels.
Rethinking memory in continual learning: Beyond a monolithic store of the past
(TMLR, 2025) Zhang, Yaqian; Pfahringer, Bernhard; Frank, Eibe; Bifet, Albert
Memory is a critical component in replay-based continual learning (CL). Prior research has largely treated CL memory as a monolithic store of past data, focusing on how to select and store representative past examples. However, this perspective overlooks the higher-level memory architecture that governs the interaction between old and new data. In this work, we identify and characterize a dual-memory system that is inherently present in both online and offline CL settings. This system comprises: a short-term memory, which temporarily buffers recent data for immediate model updates, and a long-term memory, which maintains a carefully curated subset of past experiences for future replay and consolidation. We propose memory capacity ratio (MCR), the ratio between short-term memory and long-term memory capacities, to characterize online and offline CL. Based on this framework, we systematically investigate how MCR influences generalization, stability, and plasticity. Across diverse CL settings—class-incremental, task-incremental, and domain-incremental—and multiple data modalities (e.g., image and text classification), we observe that a smaller MCR, characteristic of online CL, can yield comparable or even superior performance relative to a larger one, characteristic of offline CL, when both are evaluated under equivalent computational and data storage budgets. This advantage holds consistently across several state-of-the-art replay strategies, such as ER, DER, and SCR. Theoretical analysis further reveals that a reduced MCR yields a better trade-off between stability and plasticity by lowering a bound on generalization error when learning from non-stationary data streams with limited memory. These findings offer new insights into the role of memory allocation in continual learning and underscore the underexplored potential of online CL approaches.
BELLA: Black-box model explanations by local linear approximations
(TMLR, 2025) Radulović, Nedeljko; Bifet, Albert; Suchanek, Fabian M.
Understanding the decision-making process of black-box models has become not just a legal requirement, but also an additional way to assess their performance. However, the state of the art post-hoc explanation approaches for regression models rely on synthetic data generation, which introduces uncertainty and can hurt the reliability of the explanations. Furthermore, they tend to produce explanations that apply to only very few data points. In this paper, we present BELLA, a deterministic model-agnostic post-hoc approach for explaining the individual predictions of regression black-box models. BELLA provides explanations in the form of a linear model trained in the feature space. BELLA maximizes the size of the neighborhood to which the linear model applies so that the explanations are accurate, simple, general, and robust.
Machine learning for data streams with CapyMOA
(Springer, 2026) Sun, Yibin; Gomes, Heitor Murilo; Lee, Anton; Gunasekara, Nuwan; Cassales, Guilherme; Liu, Justin Jia; Heyden, Marco; Cerqueira, Vitor; Bahri, Maroua; Koh, Yun Sing; Pfahringer, Bernhard; Bifet, Albert
The exponential growth of data in recent decades has underscored the need for high-speed, real-time, and adaptive processing in machine learning. Data stream learning provides an effective framework to address this challenge. This article introduces CapyMOA, an open-source library designed specifically for data stream learning, offering powerful tools for building and deploying adaptive ML models. GitHub: https://github.com/adaptive-machine-learning/CapyMOA. Website: https://capymoa.org.
Streaming Isolation Forest
(Springer, 2025) Liu, Justin Jia; Cassales, Guilherme; Liu, Fei T.; Pfahringer, Bernhard; Bifet, Albert
Anomaly detection is crucial to identify unusual patterns in various domains. In particular, continuous and rapid flow creates distinct challenges within streaming data. This paper introduces the Streaming Isolation Forest (SiForest), a novel algorithm that uses isolation principles and reservoir sampling to align the model with current data distributions. SiForest efficiently detects anomalies with minimal computational and memory requirements and dynamically updates its model using a subtree regrowing strategy. Empirical evaluation on twenty-three benchmark datasets demonstrates that SiForest outperforms eight state-of-the-art algorithms in terms of AUC-ROC scores, achieving greater precision and adaptability.
Adaptive Isolation Forest
(Springer, 2025) Liu, Justin Jia; Cassales, Guilherme; Liu, Fei T.; Pfahringer, Bernhard; Bifet, Albert
Anomaly detection in real-world data streams often struggles with concept drift, where evolving data distributions challenge algorithms to maintain a balance between accuracy, speed, stability, and plasticity. We present Adaptive Isolation Forest (AIF), a novel anomaly detection algorithm designed to effectively adapt to such changes in a resource-efficient and balanced manner. AIF’s novelty can be found in the combination of a smart model update mechanism together with a newly developed MinTreeMaxMass (MTMM) criterion, which scores individual trees for replacement. Extensive evaluations on various benchmark datasets demonstrate that AIF significantly outperforms existing state-of-the-art streaming anomaly detection algorithms in terms of detection accuracy. Moreover, AIF achieves linear time and space complexities, providing a robust solution that maintains high accuracy and efficiency, balancing stability and plasticity in dynamic data streams.
Detection of human and machine-authored fake news in Urdu
(Association for Computational Linguistics, 2025) Ali, Muhammad Zain; Wang, Yuxia; Pfahringer, Bernhard; Smith, Tony C.
The rise of social media has amplified the spread of fake news, now further complicated by large language models (LLMs) like Chat-GPT, which ease the generation of highly convincing, error-free misinformation, making it increasingly challenging for the public to discern truth from falsehood. Traditional fake news detection methods relying on linguistic cues have also become less effective. Moreover, current detectors primarily focus on binary classification and English texts, often overlooking the distinction between machine-generated true vs. fake news and the detection in low-resource languages. To this end, we updated the detection schema to include machine-generated news focusing on Urdu. We further propose a conjoint detection strategy to improve the accuracy and robustness. Experiments show its effectiveness across four datasets in various settings.
Supporting interactive system testing with interaction sequences
(ACM, 2017) Turner, Jessica; Bowen, Judy; Reeves, Steve
In software engineering testing is an important part of the development process. In interactive systems human input introduces the possibility of human error, which increases the testing requirements. In safety-critical systems user or system error can result in injury or death to a user. We propose using interaction sequences as a way of supporting interactive system testing to help address these issues.
Towards formalisation of 3D user interaction: Making spatial interaction verifiable
(ACM, 2025) Turner, Jessica; Weyers, Benjamin
Interactive systems which incorporate 3D interaction techniques, such as virtual or augmented reality, lack appropriate methods for formalisation to verify and validate their properties. In safety-critical contexts, such as virtual environments used for training users in medical settings, ensuring systems behave as expected is crucial to enabling positive user experiences. In this paper we propose a formalisation that will allow us to express 3D virtual environments, end users and their interactions. This work represents a first step towards creating formalisation methods for virtual reality applications, enabling model checking techniques to ensure systems behave as expected, leading to improved reliability, safety and better user experience in virtual environments.
How positioning wearable haptic interfaces on limbs influences virtual embodiment
(IEEE, 2024) Dwivedi, Anany; Yu, Shihan; Hao, Chenxu; Salvietti, Gionata; Prattichizzo, Domenico; Beckerle, Philipp
With increasing use of computer applications and robotic devices in our everyday life, and with the advent of metaverse, there is an urgent need of developing new types of interfaces that facilitate a more intuitive interaction in physical and virtual space. In this work, we investigate the influence of the location of haptic feedback devices on embodiment of virtual hands and user load during an interactive pick-and-place task. To do this, we conducted a user study with a 3x2 repeated measure experiment design: feedback position is varied between the distal phalanx of the index finger and the thumb, the proximal phalanx of the index finger and the thumb, and the wrist. These conditions of feedback are tested with the stimuli applied synchronously to the participant in one case, and with an additional delay of 350 ms in the second case. The results show that the location of the haptic feedback device does not affect embodiment, whereas the delay, i.e., whether the feedback is applied synchronously or asynchronously, affects embodiment. This suggests that for pick-and-place tasks, haptic feedback devices can be placed on the user's wrist without compromising performance making the hands to remain free, allowing unobstructed hand visibility for precise motion tracking, thereby improving accuracy.
On the impact of different light wavelengths in decoding human intention in lightmyography controlled prosthetic hands
(IEEE, 2025) Guan, Bonnie; Godoy, Ricardo V.; Dwivedi, Anany; Liarokapis, Minas
Lightmyography is a method of human interfacing proposed for the control of prosthetic systems in which light is reflected on the skin and captured using photosensors. While sharing conceptual similarities with forcemyography, lightmyography differentiates itself in the way it interacts with the skin due to skin optics. Different wavelengths of light interact with the skin in different depths and degrees, a principle utilized in photoplethysmography to estimate heart rate and blood oxygen levels. Therefore, lightmyography was previously designed with two wavelengths of light, green and near-infrared. This study seeks to examine the impact of using different wavelengths of light in the decoding of hand postures using different machine learning methods. Four light configurations were tested: using only green light, using only infrared light, using both green and infrared light blinking alternately each 250 ms, and finally using both lights blinking at 125 ms. Data was decoded using four machine learning algorithms, namely linear discriminant analysis, random forest, convolutional neural networks, and a transformer-based model. Statistical testing showed no significant effect from light configuration in the machine learning decoding of hand posture performance. The dataset collected in this study has been made available for further investigation.
Choosing the right time to learn evolving data streams
(IEEE, 2023) Bernardo, Alessio; Della Valle, Emanuele; Bifet, Albert
Continuous data generation over time presents new challenges for Machine Learning systems, which must develop real-time models due to memory and latency limitations. Streaming Machine Learning algorithms analyze data streams one sample at a time, progressively updating their models. However, is it necessary to utilize all the data for model updates? This paper introduces the Online Ensemble SPaced Learning (OE-SPL) strategy, an ensemble meta-strategy that combines online ensemble learning and the Spaced Learning heuristic to rapidly learn underlying concepts without using all samples. We evaluated OE-SPL on synthetic and real data streams containing various concept drifts, providing statistical evidence that OE-SPL achieves comparable performance to state-of-the-art ensemble models while recovering from multiple concept drift occurrences more efficiently, using less time and RAM-Hours.
Defining sets which intersect each Latin trade at least twice
(Elsevier, 2026) Bean, Richard; Cavenagh, Nicholas J.
A defining set of a Latin square is a partially filled-in Latin square which completes to no other Latin square of the same order. We introduce the concept of a k -strong defining set, in which if less than k entries are deleted, the property of being a defining set is retained. Equivalently, a k -strong defining set intersects every Latin trade in the Latin square at least k times. In the addition table for integers modulo n , when n is even we determine the minimum size of a k -strong defining set for any k . For odd n we give a construction for a minimally 2-strong defining set. We furthermore give computational results for Latin squares of small orders.
Social capital is critical to perseverance at university during COVID-19: Pasifika students in Fiji, New Zealand and Solomon Islands universities
(University of Sydney, 2022) Dorovolomo, Jeremy; Laulaupea’alu, Siuta; Rafiq, Loriza Zinnie; Rodie, Patricia; Fito’o, Billy
This study investigated how university students in two Pacific Island contexts coped with the shift to remote learning caused by the COVID-19 pandemic. The researchers aimed to investigate the social capital of a group of Pasifika university students in New Zealand and the Solomon Islands based in Fiji and the Solomon Islands during COVID-19 and how they coped with the rapid shift to remote learning. Social capital refers to building relationships and networks to deal with collective issues in the present and future. The researchers used tokstori to collect data from Solomon Islands students and talanoa in New Zealand. Sixteen Solomon Islanders participated in tokstoris and ongoing Pasifika students’ talanoa and faikava (kava consumption) to share concerns and challenges to their academic studies and lives. Deductive in nature, the analysis found that communication with fellow students, often via digital media, and interaction with supportive staff helped their success. Furthermore, getting in touch with the family and following COVID-19 protective practices helped students adjust positively. Being able to ‘tap’ the social resources within these spaces helped build perseverance, resilience and strong social capital. The research findings show the importance of individuals, student communities and institutions consciously strengthening social networks among students during a crisis such as the COVID-19 pandemic.
Viability of automating the landing error scoring system using inertial measurement units
(Elsevier, 2026) Hamilton, Zane; Turner, Jessica; König, Jemma; Hébert-Losier, Kim
The Landing Error Scoring System (LESS) is an assessment tool used for identifying movement patterns linked with non-contact anterior cruciate ligament injuries during a double-leg jump-landing; however, the LESS is scored by experts using 2D video recordings, which limits large-scale screening. This study explores the viability of using inertial measurement unit (IMU) data to automate scoring of 17 LESS items, using conventional machine learning model architecture. Forty healthy participants completed six jumps each, and raw movement data from three IMU sensors placed on the sacrum and medial-inferior aspects of the tibias were processed and segmented into the key phases of the double-leg jump-landing. A total of 218 jumps were used to train supervised machine learning models using various subsets of the processed dataset, including extracted temporal and statistical features. Performance metrics were extracted, and the best performing models for each scoring item were evaluated against a majority-class baseline classifier (ZeroR). Results indicated moderate improvements over ZeroR, ranging from 0.3 % to 22.5 %, with some models demonstrating limited recognition of the minority class. These findings provide a foundation for future research in IMU-based LESS automation, and improving the accessibility of movement screening tools.

Browse

Recent Submissions