arXiv subjects

Michael Stewart

Publications and source records attributed to Michael Stewart.

At least 19 recordsLinked to original sources

arXiv2026

Cognitive Fatigue in Autoregressive Transformers: Formalization and Measurement

Autoregressive language models frequently degrade during long-horizon generation, producing repetitive text, losing instruction adherence, and exhibiting unstable entropy. Despite the prevalence of these failures, practitioners lack online diagnostics to detect them in real-time as they occur. We formalize this degradation as cognitive fatigue, a measurable generation-time state characterized by decay in attention to the original prompt, representational drift, and entropy miscalibration. We introduce the Fatigue Index (FI), a lightweight, model-agnostic diagnostic that aggregates these three signals under explicit axioms (monotonicity, boundedness, interpretability) enabling reliable runtime monitoring. Across nine models (1B-13B parameters), FI trajectories exhibit structured temporal dynamics, predict task degradation (AUROC = 0.95) and repetition (Spearman rho = 0.94), and reveal non-monotonic scaling behavior: instruction-tuned models below 3B exhibit faster collapse than base models, with this trend reversing at 7B. Stress analyses further show that FI onset accelerates under longer contexts, middle-positioned evidence, and reduced numerical precision. These results establish cognitive fatigue as a coherent and measurable phenomenon, and position FI as a principled tool for runtime reliability monitoring in production LLM systems.

cs.CL↗

arXiv2024

Spectral Transformation for the Dense Symmetric Semidefinite Generalized Eigenvalue Problem

The spectral transformation Lanczos method for the sparse symmetric definite generalized eigenvalue problem for matrices $A$ and $B$ is an iterative method that addresses the case of semidefinite or ill conditioned $B$ using a shifted and inverted formulation of the problem. This paper proposes the same approach for dense problems and shows that with a shift chosen in accordance with certain constraints, the algorithm can conditionally ensure that every computed shifted and inverted eigenvalue is close to the exact shifted and inverted eigenvalue of a pair of matrices close to $A$ and $B$. Under the same assumptions on the shift, the analysis of the algorithm for the shifted and inverted problem leads to useful error bounds for the original problem, including a bound that shows how a single shift that is of moderate size in a scaled sense can be chosen so that every computed generalized eigenvalue corresponds to a generalized eigenvalue of a pair of matrices close to $A$ and $B$. The computed generalized eigenvectors give a relative residual that depends on the distance between the corresponding generalized eigenvalue and the shift. If the shift is of moderate size, then relative residuals are small for generalized eigenvalues that are not much larger than the shift. Larger shifts give small relative residuals for generalized eigenvalues that are not much larger or smaller than the shift.

math.NA↗

arXiv2024

CleanGraph: Human-in-the-loop Knowledge Graph Refinement and Completion

This paper presents CleanGraph, an interactive web-based tool designed to facilitate the refinement and completion of knowledge graphs. Maintaining the reliability of knowledge graphs, which are grounded in high-quality and error-free facts, is crucial for real-world applications such as question-answering and information retrieval systems. These graphs are often automatically assembled from textual sources by extracting semantic triples via information extraction. However, assuring the quality of these extracted triples, especially when dealing with large or low-quality datasets, can pose a significant challenge and adversely affect the performance of downstream applications. CleanGraph allows users to perform Create, Read, Update, and Delete (CRUD) operations on their graphs, as well as apply models in the form of plugins for graph refinement and completion tasks. These functionalities enable users to enhance the integrity and reliability of their graph data. A demonstration of CleanGraph and its source code can be accessed at https://github.com/nlp-tlp/CleanGraph under the MIT License.

cs.AI↗

arXiv2023

Rel2Graph: Automated Mapping From Relational Databases to a Unified Property Knowledge Graph

Although a few approaches are proposed to convert relational databases to graphs, there is a genuine lack of systematic evaluation across a wider spectrum of databases. Recognising the important issue of query mapping, this paper proposes an approach Rel2Graph, an automatic knowledge graph construction (KGC) approach from an arbitrary number of relational databases. Our approach also supports the mapping of conjunctive SQL queries into pattern-based NoSQL queries. We evaluate our proposed approach on two widely used relational database-oriented datasets: Spider and KaggleDBQA benchmarks for semantic parsing. We employ the execution accuracy (EA) metric to quantify the proportion of results by executing the NoSQL queries on the property knowledge graph we construct that aligns with the results of SQL queries performed on relational databases. Consequently, the counterpart property knowledge graph of benchmarks with high accuracy and integrity can be ensured. The code and data will be publicly available. The code and data are available at github\footnote{https://github.com/nlp-tlp/Rel2Graph}.

cs.DB↗

arXiv2023

Large Language Models for Failure Mode Classification: An Investigation

In this paper we present the first investigation into the effectiveness of Large Language Models (LLMs) for Failure Mode Classification (FMC). FMC, the task of automatically labelling an observation with a corresponding failure mode code, is a critical task in the maintenance domain as it reduces the need for reliability engineers to spend their time manually analysing work orders. We detail our approach to prompt engineering to enable an LLM to predict the failure mode of a given observation using a restricted code list. We demonstrate that the performance of a GPT-3.5 model (F1=0.80) fine-tuned on annotated data is a significant improvement over a currently available text classification model (F1=0.60) trained on the same annotated data set. The fine-tuned model also outperforms the out-of-the box GPT-3.5 (F1=0.46). This investigation reinforces the need for high quality fine-tuning data sets for domain-specific tasks using LLMs.

cs.CL↗

arXiv2022

Standing Balance Improvement Using Vibrotactile Feedback in Virtual Reality

Virtual Reality (VR) users often encounter postural instability, i.e., balance issues, which can be a significant impediment to universal usability and accessibility, particularly for those with balance impairments. Prior research has validated imbalance issues, but little effort has been made to mitigate them. We recruited 39 participants (with balance impairments: 18, without balance impairments: 21) to examine the effect of various vibrotactile feedback techniques on balance in virtual reality, specifically spatial vibrotactile, static vibrotactile, rhythmic vibrotactile, and vibrotactile feedback mapped to the center of pressure (CoP). Participants completed standing visual exploration and standing reach and grasp tasks. According to within-subject results, each vibrotactile feedback enhanced balance in VR significantly (p < .001) for those with and without balance impairments. Spatial and CoP vibrotactile feedback enhanced balance significantly more (p < .001) than other vibrotactile feedback. This study presents strategies that might be used in future virtual environments to enhance standing balance and bring VR closer to universal usage.

cs.HC↗

arXiv2022

Auditory Feedback to Make Walking in Virtual Reality More Accessible

The objective of this study is to investigate the impact of several auditory feedback modalities on gait (i.e., walking patterns) in virtual reality (VR). Prior research has substantiated gait disturbances in VR users as one of the primary obstacles to VR usability. However, minimal research has been done to mitigate this issue. We recruited 39 participants (with mobility impairments: 18, without mobility impairments: 21) who completed timed walking tasks in a real-world environment and the same tasks in a VR environment with various types of auditory feedback. Within-subject results showed that each auditory condition significantly improved gait performance while in VR (p < .001) compared to the no auditory condition in VR for both groups of participants with and without mobility impairments. Moreover, spatial audio improved gait performance significantly (p < .001) compared to other auditory conditions for both groups of participants. This research could help to make walking in VR more accessible for people with and without mobility impairments.

cs.HC↗

arXiv2022

Vibrotactile Feedback to Make Real Walking in Virtual Reality More Accessible

This research aims to examine the effects of various vibrotactile feedback techniques on gait (i.e., walking patterns) in virtual reality (VR). Prior studies have demonstrated that gait disturbances in VR users are significant usability barriers. However, adequate research has not been performed to address this problem. In our study, 39 participants (with mobility impairments: 18, without mobility impairments: 21) performed timed walking tasks in a real-world environment and identical activities in a VR environment with different forms of vibrotactile feedback (spatial, static, and rhythmic). Within-group results revealed that each form of vibrotactile feedback improved gait performance in VR significantly (p < .001) relative to the no vibrotactile condition in VR for individuals with and without mobility impairments. Moreover, spatial vibrotactile feedback increased gait performance significantly (p < .001) in both participant groups compared to other vibrotactile conditions. The findings of this research will help to make real walking in VR more accessible for those with and without mobility impairments.

cs.HC↗

arXiv2022

Auditory Feedback for Standing Balance Improvement in Virtual Reality

Virtual Reality (VR) users often experience postural instability, i.e., balance problems, which could be a major barrier to universal usability and accessibility for all, especially for persons with balance impairments. Prior research has confirmed the imbalance effect, but minimal research has been conducted to reduce this effect. We recruited 42 participants (with balance impairments: 21, without balance impairments: 21) to investigate the impact of several auditory techniques on balance in VR, specifically spatial audio, static rest frame audio, rhythmic audio, and audio mapped to the center of pressure (CoP). Participants performed two types of tasks - standing visual exploration and standing reach and grasp. Within-subject results showed that each auditory technique improved balance in VR for both persons with and without balance impairments. Spatial and CoP audio improved balance significantly more than other auditory conditions. The techniques presented in this research could be used in future virtual environments to improve standing balance and help push VR closer to universal usability.

cs.HC↗

arXiv2020

E2EET: From Pipeline to End-to-end Entity Typing via Transformer-Based Embeddings

Entity Typing (ET) is the process of identifying the semantic types of every entity within a corpus. In contrast to Named Entity Recognition, where each token in a sentence is labelled with zero or one class label, ET involves labelling each entity mention with one or more class labels. Existing entity typing models, which operate at the mention level, are limited by two key factors: they do not make use of recently-proposed context-dependent embeddings, and are trained on fixed context windows. They are therefore sensitive to window size selection and are unable to incorporate the context of the entire document. In light of these drawbacks we propose to incorporate context using transformer-based embeddings for a mention-level model, and an end-to-end model using a Bi-GRU to remove the dependency on window size. An extensive ablative study demonstrates the effectiveness of contextualised embeddings for mention-level models and the competitiveness of our end-to-end model for entity typing.

cs.CL↗

arXiv2020

Dynamics of Matter-Wave Quantum Emitters in a Structured Vacuum

The characteristics of spontaneous emission can be strongly modified by the mode structure of the vacuum. In waveguide quantum-electrodynamics based on photonic crystals, this modification is exploited to engineer atom-photon interactions near a band edge, but the physics of coupling to an entire band has not yet been explored in experiments. Using ultracold atoms in an optical lattice, we study the decay dynamics of matter-wave quantum emitters coupled to a single band of an effective photonic crystal waveguide structure with tunable characteristics. Depending on the ratio between vacuum coupling and bandwidth, we observe a transition from irreversible decay to fully oscillatory dynamics linked to the interplay of matter-wave bound states near the band edges, whose spatial structure we characterize. Our results shed light on the emergence of coherence in an open quantum system in a controllable environment, and are of relevance for the understanding of vacuum-induced decay phenomena in photonic systems.

cond-mat.quant-gas↗

arXiv2019

Word-level Lexical Normalisation using Context-Dependent Embeddings

Lexical normalisation (LN) is the process of correcting each word in a dataset to its canonical form so that it may be more easily and more accurately analysed. Most lexical normalisation systems operate at the character-level, while word-level models are seldom used. Recent language models offer solutions to the drawbacks of word-level LN models, yet, to the best of our knowledge, no research has investigated their effectiveness on LN. In this paper we introduce a word-level GRU-based LN model and investigate the effectiveness of recent embedding techniques on word-level LN. Our results show that our GRU-based word-level model produces greater results than character-level models, and outperforms existing deep-learning based LN techniques on Twitter data. We also find that randomly-initialised embeddings are capable of outperforming pre-trained embedding models in certain scenarios. Finally, we release a substantial lexical normalisation dataset to the community.

cs.CL↗

arXiv2019

ICDM 2019 Knowledge Graph Contest: Team UWA

We present an overview of our triple extraction system for the ICDM 2019 Knowledge Graph Contest. Our system uses a pipeline-based approach to extract a set of triples from a given document. It offers a simple and effective solution to the challenge of knowledge graph construction from domain-specific text. It also provides the facility to visualise useful information about each triple such as the degree, betweenness, structured relation type(s), and named entity types.

cs.CL↗

arXiv2019

Variational Discriminant Analysis with Variable Selection

A fast Bayesian method that seamlessly fuses classification and hypothesis testing via discriminant analysis is developed. Building upon the original discriminant analysis classifier, modelling components are added to identify discriminative variables. A combination of cake priors and a novel form of variational Bayes we call reverse collapsed variational Bayes gives rise to variable selection that can be directly posed as a multiple hypothesis testing approach using likelihood ratio statistics. Some theoretical arguments are presented showing that Chernoff-consistency (asymptotically zero type I and type II error) is maintained across all hypotheses. We apply our method on some publicly available genomics datasets and show that our method performs well in practice for its computational cost. An R package VaDA has also been made available on Github.

stat.ME↗

arXiv2017

Spontaneous Emission in a Matter-Wave Open Quantum System

One of the paradigms of a small quantum system in a dissipative environment is the decay of an excited atom undergoing spontaneous photon emission into the fluctuating quantum electrodynamic vacuum. Recent experiments have demonstrated that the gapped photon dispersion in periodic structures can give rise to novel spontaneous-decay behavior including the formation of dissipative bound states. So far, these effects have been restricted to the optical domain. Here, we experimentally demonstrate similar behavior in a system of artificial atoms in an optical lattice that decay by emitting matter-wave, rather than optical, radiation into free space. By controlling the vacuum coupling and excitation energy, we directly observe exponential and partly reversible, non-Markovian dynamics and detect a tunable bound state containing evanescent matter waves for emission at negative excitation energies. Our system provides a flexible platform for the emulation of open-system quantum electrodynamics and studies of dissipative many-body physics with ultracold atoms.

cond-mat.quant-gas↗

arXiv2017

Bayesian hypothesis tests with diffuse priors: Can we have our cake and eat it too?

We introduce a new class of priors for Bayesian hypothesis testing, which we name "cake priors". These priors circumvent Bartlett's paradox (also called the Jeffreys-Lindley paradox); the problem associated with the use of diffuse priors leading to nonsensical statistical inferences. Cake priors allow the use of diffuse priors (having one's cake) while achieving theoretically justified inferences (eating it too). We demonstrate this methodology for Bayesian hypotheses tests for scenarios under which the one and two sample t-tests, and linear models are typically derived. The resulting Bayesian test statistic takes the form of a penalized likelihood ratio test statistic. By considering the sampling distribution under the null and alternative hypotheses we show for independent identically distributed regular parametric models that Bayesian hypothesis tests using cake priors are Chernoff-consistent, i.e., achieve zero type I and II errors asymptotically. Lindley's paradox is also discussed. We argue that a true Lindley's paradox will only occur with small probability for large sample sizes.

math.ST↗

arXiv2017

Analysis of non-Markovian coupling of a lattice-trapped atom to free space

Behavior analogous to that of spontaneous emission in photonic band gap materials has been predicted for an atom-optical system consisting of an atom confined in a well of a state-dependent optical lattice that is coupled to free space through an internal-state transition [de Vega et al., Phys. Rev. Lett. 101, 260404 (2008)]. Using the Weisskopf-Wigner approach and considering a one-dimensional geometry, we analyze the properties of this system in detail, including the evolution of the lattice-trapped population, the momentum distribution of emitted matter waves, and the detailed structure of an evanescent matter-wave state below the continuum boundary. We compare and contrast our findings for the transition from Markovian to non-Markovian behaviors to those previously obtained for three dimensions.

cond-mat.quant-gas↗

arXiv2017

In-situ magnetometry for experiments with atomic quantum gases

Precise control of magnetic fields is a frequent challenge encountered in experiments with atomic quantum gases. Here we present a simple method for performing in-situ monitoring of magnetic fields that can readily be implemented in any quantum-gas apparatus in which a dedicated field-stabilization approach is not possible. The method, which works by sampling several Rabi resonances between magnetically field sensitive internal states that are not otherwise used in a given experiment, can be integrated with standard measurement sequences at arbitrary fields. For a condensate of $^{87}$Rb atoms, we demonstrate the reconstruction of Gauss-level bias fields with an accuracy of tens of microgauss and with millisecond time resolution. We test the performance of the method using measurements of slow resonant Rabi oscillations on a magnetic-field sensitive transition, and give an example for its use in experiments with state-selective optical potentials.

cond-mat.quant-gas↗