Raul Santos-Rodriguez

Kacper Sokol, Alexander Hepburn, Rafael Poyiadzi et al.

Predictive systems, in particular machine learning algorithms, can take important, and sometimes legally binding, decisions about our everyday life. In most cases, however, these systems and decisions are neither regulated nor certified. Given the potential harm that these algorithms can cause, their qualities such as fairness, accountability and transparency (FAT) are of paramount importance. To ensure high-quality, fair, transparent and reliable predictive systems, we developed an open source Python package called FAT Forensics. It can inspect important fairness, accountability and transparency aspects of predictive algorithms to automatically and objectively report them back to engineers and users of such systems. Our toolbox can evaluate all elements of a predictive pipeline: data (and their features), models and predictions. Published under the BSD 3-Clause open source licence, FAT Forensics is opened up for personal and commercial usage.

Kacper Sokol, Alexander Hepburn, Raul Santos-Rodriguez et al.

Explainability techniques for data-driven predictive models based on artificial intelligence and machine learning algorithms allow us to better understand the operation of such systems and help to hold them accountable. New transparency approaches are developed at breakneck speed, enabling us to peek inside these black boxes and interpret their decisions. Many of these techniques are introduced as monolithic tools, giving the impression of one-size-fits-all and end-to-end algorithms with limited customisability. Nevertheless, such approaches are often composed of multiple interchangeable modules that need to be tuned to the problem at hand to produce meaningful explanations. This paper introduces a collection of hands-on training materials -- slides, video recordings and Jupyter Notebooks -- that provide guidance through the process of building and evaluating bespoke modular surrogate explainers for tabular data. These resources cover the three core building blocks of this technique: interpretable representation composition, data sampling and explanation generation.

Armand K. Koupai, Mohammud J. Bocus, Raul Santos-Rodriguez et al.

The pervasiveness of Wi-Fi signals provides significant opportunities for human sensing and activity recognition in fields such as healthcare. The sensors most commonly used for passive Wi-Fi sensing are based on passive Wi-Fi radar (PWR) and channel state information (CSI) data, however current systems do not effectively exploit the information acquired through multiple sensors to recognise the different activities. In this paper, we explore new properties of the Transformer architecture for multimodal sensor fusion. We study different signal processing techniques to extract multiple image-based features from PWR and CSI data such as spectrograms, scalograms and Markov transition field (MTF). We first propose the Fusion Transformer, an attention-based model for multimodal and multi-sensor fusion. Experimental results show that our Fusion Transformer approach can achieve competitive results compared to a ResNet architecture but with much fewer resources. To further improve our model, we propose a simple and effective framework for multimodal and multi-sensor self-supervised learning (SSL). The self-supervised Fusion Transformer outperforms the baselines, achieving a F1-score of 95.9%. Finally, we show how this approach significantly outperforms the others when trained with as little as 1% (2 minutes) of labelled training data to 20% (40 minutes) of labelled training data.

Yuxuan Ye, Raul Santos-Rodriguez, Edwin Simpson

Grounded claim factuality checking is important for large language model (LLM) applications such as retrieval-augmented generation, as it helps users assess the correctness of generated outputs. Existing metrics using entailment classifiers require dataset-specific threshold tuning, while LLM-based approaches often use direct prompting, which underutilises the reasoning capabilities of LLMs. We address this by formulating grounded claim factuality checking as a true/false reading comprehension task and prompting LLMs with explicit test-taking strategies for efficient reasoning. Our method reduces token usage by over 80% compared to unguided open-ended reasoning, and achieves competitive performance to more expensive alternatives across two factuality benchmarks, setting a new state of the art on one. To further reduce inference cost, we train small language models (SLMs) to replace LLMs in the checking pipeline. Using supervised fine-tuning (SFT) and a self-revision mechanism, the SLMs learn to improve their factuality judgements. Experimental results show that the resulting SLMs perform on par with strong baselines, combining low inference costs with generating supporting rationales to support interpretability. Code and datasets will be released upon acceptance.

Taku Yamagata, Emma L. Tonkin, Benjamin Arana Sanchez et al.

In supervised learning, low quality annotations lead to poorly performing classification and detection models, while also rendering evaluation unreliable. This is particularly apparent on temporal data, where annotation quality is affected by multiple factors. For example, in the post-hoc self-reporting of daily activities, cognitive biases are one of the most common ingredients. In particular, reporting the start and duration of an activity after its finalisation may incorporate biases introduced by personal time perceptions, as well as the imprecision and lack of granularity due to time rounding. Here we propose a method to model human biases on temporal annotations and argue for the use of soft labels. Experimental results in synthetic data show that soft labels provide a better approximation of the ground truth for several metrics. We showcase the method on a real dataset of daily activities.

Ahmed Khalil, Robert Piechocki, Raul Santos-Rodriguez

In this paper we introduce learnable lattice vector quantization and demonstrate its effectiveness for learning discrete representations. Our method, termed LL-VQ-VAE, replaces the vector quantization layer in VQ-VAE with lattice-based discretization. The learnable lattice imposes a structure over all discrete embeddings, acting as a deterrent against codebook collapse, leading to high codebook utilization. Compared to VQ-VAE, our method obtains lower reconstruction errors under the same training conditions, trains in a fraction of the time, and with a constant number of parameters (equal to the embedding dimension $D$), making it a very scalable approach. We demonstrate these results on the FFHQ-1024 dataset and include FashionMNIST and Celeb-A.

Yuxuan Ye, Raul Santos-Rodriguez, Edwin Simpson

Reinforcement learning with evaluation metrics as rewards is widely used to enhance specific capabilities of language models. However, for tasks such as factually consistent summarisation, existing metrics remain underdeveloped, limiting their effectiveness as signals for shaping model behaviour.While individual factuality metrics are unreliable, their combination can more effectively capture diverse factual errors. We leverage this insight to introduce an automated training pipeline that improves factual consistency in summaries by aggregating scores from different weak metrics. Our approach avoids the need for complex reward shaping by mapping scores to preferences and filtering out cases with high disagreement between metrics. For each source document, we generate lexically similar summary pairs by varying decoding strategies, enabling the model to learn from factual differences caused by subtle lexical differences. This approach constructs a high-quality preference dataset using only source documents.Experiments demonstrate consistent factuality gains across models, ranging from early encoder-decoder architectures to modern large language models, with smaller models reaching comparable factuality to larger ones.

Taku Yamagata, Ahmed Khalil, Raul Santos-Rodriguez

Recent works have shown that tackling offline reinforcement learning (RL) with a conditional policy produces promising results. The Decision Transformer (DT) combines the conditional policy approach and a transformer architecture, showing competitive performance against several benchmarks. However, DT lacks stitching ability -- one of the critical abilities for offline RL to learn the optimal policy from sub-optimal trajectories. This issue becomes particularly significant when the offline dataset only contains sub-optimal trajectories. On the other hand, the conventional RL approaches based on Dynamic Programming (such as Q-learning) do not have the same limitation; however, they suffer from unstable learning behaviours, especially when they rely on function approximation in an off-policy learning setting. In this paper, we propose the Q-learning Decision Transformer (QDT) to address the shortcomings of DT by leveraging the benefits of Dynamic Programming (Q-learning). It utilises the Dynamic Programming results to relabel the return-to-go in the training data to then train the DT with the relabelled data. Our approach efficiently exploits the benefits of these two approaches and compensates for each other's shortcomings to achieve better performance. We empirically show these in both simple toy environments and the more complex D4RL benchmark, showing competitive performance gains.

Amarpal Sahota, Amber Roguski, Matthew W. Jones et al.

Detecting Parkinson's Disease in its early stages using EEG data presents a significant challenge. This paper introduces a novel approach, representing EEG data as a 15-variate series of bandpower and peak frequency values/coefficients. The hypothesis is that this representation captures essential information from the noisy EEG signal, improving disease detection. Statistical features extracted from this representation are utilised as input for interpretable machine learning models, specifically Decision Tree and AdaBoost classifiers. Our classification pipeline is deployed within our proposed framework which enables high-importance data types and brain regions for classification to be identified. Interestingly, our analysis reveals that while there is no significant regional importance, the N1 sleep data type exhibits statistically significant predictive power (p < 0.01) for early-stage Parkinson's Disease classification. AdaBoost classifiers trained on the N1 data type consistently outperform baseline models, achieving over 80% accuracy and recall. Our classification pipeline statistically significantly outperforms baseline models indicating that the model has acquired useful information. Paired with the interpretability (ability to view feature importance's) of our pipeline this enables us to generate meaningful insights into the classification of early stage Parkinson's with our N1 models. In Future, these models could be deployed in the real world - the results presented in this paper indicate that more than 3 in 4 early-stage Parkinson's cases would be captured with our pipeline.

Edward A. Small, Jeffrey N. Clark, Christopher J. McWilliams et al.

Counterfactuals operationalised through algorithmic recourse have become a powerful tool to make artificial intelligence systems explainable. Conceptually, given an individual classified as y -- the factual -- we seek actions such that their prediction becomes the desired class y' -- the counterfactual. This process offers algorithmic recourse that is (1) easy to customise and interpret, and (2) directly aligned with the goals of each individual. However, the properties of a "good" counterfactual are still largely debated; it remains an open challenge to effectively locate a counterfactual along with its corresponding recourse. Some strategies use gradient-driven methods, but these offer no guarantees on the feasibility of the recourse and are open to adversarial attacks on carefully created manifolds. This can lead to unfairness and lack of robustness. Other methods are data-driven, which mostly addresses the feasibility problem at the expense of privacy, security and secrecy as they require access to the entire training data set. Here, we introduce LocalFACE, a model-agnostic technique that composes feasible and actionable counterfactual explanations using locally-acquired information at each step of the algorithmic recourse. Our explainer preserves the privacy of users by only leveraging data that it specifically requires to construct actionable algorithmic recourse, and protects the model by offering transparency solely in the regions deemed necessary for the intervention.

Nawid Keshtmand, Raul Santos-Rodriguez, Jonathan Lawry

A recent popular approach to out-of-distribution (OOD) detection is based on a self-supervised learning technique referred to as contrastive learning. There are two main variants of contrastive learning, namely instance and class discrimination, targeting features that can discriminate between different instances for the former, and different classes for the latter. In this paper, we aim to understand the effectiveness and limitation of existing contrastive learning methods for OOD detection. We approach this in 3 ways. First, we systematically study the performance difference between the instance discrimination and supervised contrastive learning variants in different OOD detection settings. Second, we study which in-distribution (ID) classes OOD data tend to be classified into. Finally, we study the spectral decay property of the different contrastive learning approaches and examine how it correlates with OOD detection performance. In scenarios where the ID and OOD datasets are sufficiently different from one another, we see that instance discrimination, in the absence of fine-tuning, is competitive with supervised approaches in OOD detection. We see that OOD samples tend to be classified into classes that have a distribution similar to the distribution of the entire dataset. Furthermore, we show that contrastive learning learns a feature space that contains singular vectors containing several directions with a high variance which can be detrimental or beneficial to OOD detection depending on the inference approach used.

Jeffrey N. Clark, Edward A. Small, Nawid Keshtmand et al.

Counterfactual explanations, and their associated algorithmic recourse, are typically leveraged to understand, explain, and potentially alter a prediction coming from a black-box classifier. In this paper, we propose to extend the use of counterfactuals to evaluate progress in sequential decision making tasks. To this end, we introduce a model-agnostic modular framework, TraCE (Trajectory Counterfactual Explanation) scores, which is able to distill and condense progress in highly complex scenarios into a single value. We demonstrate TraCE's utility across domains by showcasing its main properties in two case studies spanning healthcare and climate change.

Amarpal Sahota, Amber Roguski, Matthew W Jones et al.

We evaluate the effectiveness of combining brain connectivity metrics with signal statistics for early stage Parkinson's Disease (PD) classification using electroencephalogram data (EEG). The data is from 5 arousal states - wakeful and four sleep stages (N1, N2, N3 and REM). Our pipeline uses an Ada Boost model for classification on a challenging early stage PD classification task with with only 30 participants (11 PD , 19 Healthy Control). Evaluating 9 brain connectivity metrics we find the best connectivity metric to be different for each arousal state with Phase Lag Index achieving the highest individual classification accuracy of 86\% on N1 data. Further to this our pipeline using regional signal statistics achieves an accuracy of 78\%, using brain connectivity only achieves an accuracy of 86\% whereas combining the two achieves a best accuracy of 91\%. This best performance is achieved on N1 data using Phase Lag Index (PLI) combined with statistics derived from the frequency characteristics of the EEG signal. This model also achieves a recall of 80 \% and precision of 96\%. Furthermore we find that on data from each arousal state, combining PLI with regional signal statistics improves classification accuracy versus using signal statistics or brain connectivity alone. Thus we conclude that combining brain connectivity statistics with regional EEG statistics is optimal for classifier performance on early stage Parkinson's. Additionally, we find outperformance of N1 EEG for classification of Parkinson's and expect this could be due to disrupted N1 sleep in PD. This should be explored in future work.

Enrico Werner, Jeffrey N. Clark, Ranjeet S. Bhamber et al.

We present a pipeline in which unsupervised machine learning techniques are used to automatically identify subtypes of hospital patients admitted between 2017 and 2021 in a large UK teaching hospital. With the use of state-of-the-art explainability techniques, the identified subtypes are interpreted and assigned clinical meaning. In parallel, clinicians assessed intra-cluster similarities and inter-cluster differences of the identified patient subtypes within the context of their clinical knowledge. By confronting the outputs of both automatic and clinician-based explanations, we aim to highlight the mutual benefit of combining machine learning techniques with clinical expertise.

Tashi Namgyal, Alexander Hepburn, Raul Santos-Rodriguez et al.

The subjective quality of natural signals can be approximated with objective perceptual metrics. Designed to approximate the perceptual behaviour of human observers, perceptual metrics often reflect structures found in natural signals and neurological pathways. Models trained with perceptual metrics as loss functions can capture perceptually meaningful features from the structures held within these metrics. We demonstrate that using features extracted from autoencoders trained with perceptual losses can improve performance on music understanding tasks, i.e. genre classification, over using these metrics directly as distances when learning a classifier. This result suggests improved generalisation to novel signals when using perceptual metrics as loss functions for representation learning.

Nawid Keshtmand, Raul Santos-Rodriguez, Jonathan Lawry

Two fundamental requirements for the deployment of machine learning models in safety-critical systems are to be able to detect out-of-distribution (OOD) data correctly and to be able to explain the prediction of the model. Although significant effort has gone into both OOD detection and explainable AI, there has been little work on explaining why a model predicts a certain data point is OOD. In this paper, we address this question by introducing the concept of an OOD counterfactual, which is a perturbed data point that iteratively moves between different OOD categories. We propose a method for generating such counterfactuals, investigate its application on synthetic and benchmark data, and compare it to several benchmark methods using a range of metrics.

Berenika Ewart-James, Matthew Wragg, Nawid Keshtmand et al.

Epilepsy is a common, chronic neurological disorder characterized by recurrent seizures caused by sudden bursts of abnormal electrical activity in the brain. Seizures can often be unpredictable, leading to uncertainty and anxiety for people with epilepsy. To address this problem, the Epilepsy UK Priority Setting Partnership identified research into seizure forecasting technology as a priority. Seizure onsets are recorded as discrete events embedded within continuously sampled physiological signals that exhibit strong circadian and multi-day rhythms. Standard modelling approaches often treat time as linear or rely on clock-time features, which may not explicitly capture the underlying physiological phase. In this paper, we examine whether seizure onsets exhibit phase preference relative to circadian rhythms derived from wearable inter-beat interval (IBI) data. As a proof-of-concept, using 176 days wearable and seizure diary data from a single patient, we extract oscillatory components via band-limited filtering and Hilbert-based phase estimation, and test for non-uniform seizure-phase alignment using circular statistics. We observe significant circadian phase concentration, while multiday bands do not show consistent or statistically significant phase clustering in this dataset. Exploratory logistic baselines indicate modest but detectable structure beyond simple clock-time effects. We argue that explicit physiological phase representations provide an interpretable bridge between continuous wearable sensing and sparse clinical events and may augment existing seizure forecasting pipelines. We discuss implications for multi-scale modelling, patient-facing interfaces, and future multi-patient validation

Maha M. Alwuthaynani, Zahraa S. Abdallah, Raul Santos-Rodriguez

Early diagnosis of Alzheimer's disease (AD) is essential in preventing the disease's progression. Therefore, detecting AD from neuroimaging data such as structural magnetic resonance imaging (sMRI) has been a topic of intense investigation in recent years. Deep learning has gained considerable attention in Alzheimer's detection. However, training a convolutional neural network from scratch is challenging since it demands more computational time and a significant amount of annotated data. By transferring knowledge learned from other image recognition tasks to medical image classification, transfer learning can provide a promising and effective solution. Irregularities in the dataset distribution present another difficulty. Class decomposition can tackle this issue by simplifying learning a dataset's class boundaries. Motivated by these approaches, this paper proposes a transfer learning method using class decomposition to detect Alzheimer's disease from sMRI images. We use two ImageNet-trained architectures: VGG19 and ResNet50, and an entropy-based technique to determine the most informative images. The proposed model achieved state-of-the-art performance in the Alzheimer's disease (AD) vs mild cognitive impairment (MCI) vs cognitively normal (CN) classification task with a 3\% increase in accuracy from what is reported in the literature.

Thea Barnes, Enrico Werner, Jeffrey N. Clark et al.

Quantifying a patient's health status provides clinicians with insight into patient risk, and the ability to better triage and manage resources. Early Warning Scores (EWS) are widely deployed to measure overall health status, and risk of adverse outcomes, in hospital patients. However, current EWS are limited both by their lack of personalisation and use of static observations. We propose a pipeline that groups intensive care unit patients by the trajectories of observations data throughout their stay as a basis for the development of personalised risk predictions. Feature importance is considered to provide model explainability. Using the MIMIC-IV dataset, six clusters were identified, capturing differences in disease codes, observations, lengths of admissions and outcomes. Applying the pipeline to data from just the first four hours of each ICU stay assigns the majority of patients to the same cluster as when the entire stay duration is considered. In-hospital mortality prediction models trained on individual clusters had higher F1 score performance in five of the six clusters when compared against the unclustered patient cohort. The pipeline could form the basis of a clinical decision support tool, working to improve the clinical characterisation of risk groups and the early detection of patient deterioration.

Kacper Sokol, Raul Santos-Rodriguez, Peter Flach

Today, artificial intelligence systems driven by machine learning algorithms can be in a position to take important, and sometimes legally binding, decisions about our everyday lives. In many cases, however, these systems and their actions are neither regulated nor certified. To help counter the potential harm that such algorithms can cause we developed an open source toolbox that can analyse selected fairness, accountability and transparency aspects of the machine learning process: data (and their features), models and predictions, allowing to automatically and objectively report them to relevant stakeholders. In this paper we describe the design, scope, usage and impact of this Python package, which is published under the 3-Clause BSD open source licence.

Ryan McConville, Raul Santos-Rodriguez, Robert J Piechocki et al.

Deep clustering has increasingly been demonstrating superiority over conventional shallow clustering algorithms. Deep clustering algorithms usually combine representation learning with deep neural networks to achieve this performance, typically optimizing a clustering and non-clustering loss. In such cases, an autoencoder is typically connected with a clustering network, and the final clustering is jointly learned by both the autoencoder and clustering network. Instead, we propose to learn an autoencoded embedding and then search this further for the underlying manifold. For simplicity, we then cluster this with a shallow clustering algorithm, rather than a deeper network. We study a number of local and global manifold learning methods on both the raw data and autoencoded embedding, concluding that UMAP in our framework is best able to find the most clusterable manifold in the embedding, suggesting local manifold learning on an autoencoded embedding is effective for discovering higher quality discovering clusters. We quantitatively show across a range of image and time-series datasets that our method has competitive performance against the latest deep clustering algorithms, including out-performing current state-of-the-art on several. We postulate that these results show a promising research direction for deep clustering. The code can be found at https://github.com/rymc/n2d

Taku Yamagata, Raul Santos-Rodriguez

Reinforcement Learning (RL) has shown remarkable success in solving relatively complex tasks, yet the deployment of RL systems in real-world scenarios poses significant challenges related to safety and robustness. This paper aims to identify and further understand those challenges thorough the exploration of the main dimensions of the safe and robust RL landscape, encompassing algorithmic, ethical, and practical considerations. We conduct a comprehensive review of methodologies and open problems that summarizes the efforts in recent years to address the inherent risks associated with RL applications. After discussing and proposing definitions for both safe and robust RL, the paper categorizes existing research works into different algorithmic approaches that enhance the safety and robustness of RL agents. We examine techniques such as uncertainty estimation, optimisation methodologies, exploration-exploitation trade-offs, and adversarial training. Environmental factors, including sim-to-real transfer and domain adaptation, are also scrutinized to understand how RL systems can adapt to diverse and dynamic surroundings. Moreover, human involvement is an integral ingredient of the analysis, acknowledging the broad set of roles that humans can take in this context. Importantly, to aid practitioners in navigating the complexities of safe and robust RL implementation, this paper introduces a practical checklist derived from the synthesized literature. The checklist encompasses critical aspects of algorithm design, training environment considerations, and ethical guidelines. It will serve as a resource for developers and policymakers alike to ensure the responsible deployment of RL systems in many application domains.

Jess Jones, Raul Santos-Rodriguez, Sabine Hauert

Vision-language models (VLMs) have demonstrated remarkable capabilities in understanding human-object interactions, but their application to robotic systems with non-humanoid morphologies remains largely unexplored. This work investigates whether VLMs can effectively infer affordances for robots with fundamentally different embodiments than humans, addressing a critical gap in the deployment of these models for diverse robotic applications. We introduce a novel hybrid dataset that combines annotated real-world robotic affordance-object relations with VLM-generated synthetic scenarios, and perform an empirical analysis of VLM performance across multiple object categories and robot morphologies, revealing significant variations in affordance inference capabilities. Our experiments demonstrate that while VLMs show promising generalisation to non-humanoid robot forms, their performance is notably inconsistent across different object domains. Critically, we identify a consistent pattern of low false positive rates but high false negative rates across all morphologies and object categories, indicating that VLMs tend toward conservative affordance predictions. Our analysis reveals that this pattern is particularly pronounced for novel tool use scenarios and unconventional object manipulations, suggesting that effective integration of VLMs in robotic systems requires complementary approaches to mitigate over-conservative behaviour while preserving the inherent safety benefits of low false positive rates.

Jeffrey N. Clark, Matthew Wragg, Emily Nielsen et al.

There is a growing need to understand how digital systems can support clinical decision-making, particularly as artificial intelligence (AI) models become increasingly complex and less human-interpretable. This complexity raises concerns about trustworthiness, impacting safe and effective adoption of such technologies. Improved understanding of decision-making processes and requirements for explanations coming from decision support tools is a vital component in providing effective explainable solutions. This is particularly relevant in the data-intensive, fast-paced environments of intensive care units (ICUs). To explore these issues, group interviews were conducted with seven ICU clinicians, representing various roles and experience levels. Thematic analysis revealed three core themes: (T1) ICU decision-making relies on a wide range of factors, (T2) the complexity of patient state is challenging for shared decision-making, and (T3) requirements and capabilities of AI decision support systems. We include design recommendations from clinical input, providing insights to inform future AI systems for intensive care.

Tashi Namgyal, Alexander Hepburn, Raul Santos-Rodriguez et al.

Perceptual metrics are traditionally used to evaluate the quality of natural signals, such as images and audio. They are designed to mimic the perceptual behaviour of human observers and usually reflect structures found in natural signals. This motivates their use as loss functions for training generative models such that models will learn to capture the structure held in the metric. We take this idea to the extreme in the audio domain by training a compressive autoencoder to reconstruct uniform noise, in lieu of natural data. We show that training with perceptual losses improves the reconstruction of spectrograms and re-synthesized audio at test time over models trained with a standard Euclidean loss. This demonstrates better generalisation to unseen natural signals when using perceptual metrics.

Yu Qin, Phillip Sloan, Raul Santos-Rodriguez et al.

Multimodal models are ubiquitous, yet existing explainability methods are often single-modal, architecture-dependent, or too computationally expensive to run at scale. We introduce LEMON (Local Explanations via Modality-aware OptimizatioN), a model-agnostic framework for local explanations of multimodal predictions. LEMON fits a single modality-aware surrogate with group-structured sparsity to produce unified explanations that disentangle modality-level contributions and feature-level attributions. The approach treats the predictor as a black box and is computationally efficient, requiring relatively few forward passes while remaining faithful under repeated perturbations. We evaluate LEMON on vision-language question answering and a clinical prediction task with image, text, and tabular inputs, comparing against representative multimodal baselines. Across backbones, LEMON achieves competitive deletion-based faithfulness while reducing black-box evaluations by 35-67 times and runtime by 2-8 times compared to strong multimodal baselines.

Lincen Bai, Hedi Tabia, Raul Santos-Rodriguez

Pruning vision-language models (VLMs) for efficient deployment is challenging because compression can affect not only task utility but also visual grounding, often amplifying object hallucinations even at the same sparsity level. We present HiPP-Prune, a hierarchical preference-conditioned structured pruning framework that treats pruning as conditional resource allocation under multiple objectives. HiPP-Prune makes plan-level decisions: a single policy invocation outputs a global pruning blueprint by factorizing decisions into an overall sparsity budget and a layer-wise allocation, enabling queryable trade-offs via a user-specified preference vector. To account for VLM-specific failure modes, our policy state integrates a visual sensitivity signal derived from attention flow between vision tokens and language hidden states, discouraging over-pruning of vision-critical layers that facilitate cross-modal fusion. We optimize pruning plans with plan-level Group Relative Policy Optimization (GRPO) under a multi-objective return that combines task utility, hallucination robustness (POPE), compression, and a synaptic-flow-inspired stability proxy to reduce unproductive exploration in high-sparsity regimes. Experiments on LLaVA with POPE and ScienceQA demonstrate that HiPP-Prune discovers diverse non-dominated pruning plans and provides controllable robustness--utility trade-offs under matched sparsity budgets.

Patricia West, Michelle WL Wan, Alexander Hepburn et al.

Climate change demands effective legislative action to mitigate its impacts. This study explores the application of machine learning (ML) to understand the progression of climate policy from announcement to adoption, focusing on policies within the European Green Deal. We present a dataset of 165 policies, incorporating text and metadata. We aim to predict a policy's progression status, and compare text representation methods, including TF-IDF, BERT, and ClimateBERT. Metadata features are included to evaluate the impact on predictive performance. On text features alone, ClimateBERT outperforms other approaches (RMSE = 0.17, R^2 = 0.29), while BERT achieves superior performance with the addition of metadata features (RMSE = 0.16, R^2 = 0.38). Using methods from explainable AI highlights the influence of factors such as policy wording and metadata including political party and country representation. These findings underscore the potential of ML tools in supporting climate policy analysis and decision-making.

Jeffrey N. Clark, Elena Fillola, Nawid Keshtmand et al.

Monitoring greenhouse gas emissions and evaluating national inventories require efficient, scalable, and reliable inference methods. Top-down approaches, combined with recent advances in satellite observations, provide new opportunities to evaluate emissions at continental and global scales. However, transport models used in these methods remain a key source of uncertainty: they are computationally expensive to run at scale, and their uncertainty is difficult to characterise. Artificial intelligence offers a dual opportunity to accelerate transport simulations and to quantify their associated uncertainty. We present an ensemble-based pipeline for estimating atmospheric transport "footprints", greenhouse gas mole fraction measurements, and their uncertainties using a graph neural network emulator of a Lagrangian Particle Dispersion Model (LPDM). The approach is demonstrated with GOSAT (Greenhouse Gases Observing Satellite) observations for Brazil in 2016. The emulator achieved a ~1000x speed-up over the NAME LPDM, while reproducing large-scale footprint structures. Ensembles were calculated to quantify absolute and relative uncertainty, revealing spatial correlations with prediction error. The results show that ensemble spread highlights low-confidence spatial and temporal predictions for both atmospheric transport footprints and methane mole fractions. While demonstrated here for an LPDM emulator, the approach could be applied more generally to atmospheric transport models, supporting uncertainty-aware greenhouse gas inversion systems and improving the robustness of satellite-based emissions monitoring. With further development, ensemble-based emulators could also help explore systematic LPDM errors, offering a computationally efficient pathway towards a more comprehensive uncertainty budget in greenhouse gas flux estimates.

Roussel Desmond Nzoyem, Nawid Keshtmand, Enrique Crespo Fernandez et al.

We introduce WARP (Weight-space Adaptive Recurrent Prediction), a simple yet powerful model that unifies weight-space learning with linear recurrence to redefine sequence modeling. Unlike conventional recurrent neural networks (RNNs) which collapse temporal dynamics into fixed-dimensional hidden states, WARP explicitly parametrizes its hidden state as the weights and biases of a distinct auxiliary neural network, and uses input differences to drive its recurrence. This brain-inspired formulation enables efficient gradient-free adaptation of the auxiliary network at test-time, in-context learning abilities, and seamless integration of domain-specific physical priors. Empirical validation shows that WARP matches or surpasses state-of-the-art baselines on diverse classification tasks, featuring in the top three in 5 out of 6 real-world challenging datasets. Furthermore, extensive experiments across sequential image completion, multivariate time series forecasting, and dynamical system reconstruction demonstrate its expressiveness and generalisation capabilities. Remarkably, a physics-informed variant of our model outperforms the next best model by more than 10x. Ablation studies confirm the architectural necessity of key components, solidifying weight-space linear RNNs as a transformative paradigm for adaptive machine intelligence.

Amarpal Sahota, Navid Mohammadi Foumani, Raul Santos-Rodriguez et al.

Electroencephalography (EEG) is a method of recording brain activity that shows significant promise in applications ranging from disease classification to emotion detection and brain-computer interfaces. Recent advances in deep learning have improved EEG classification performance yet model explainability remains an issue. To address this key limitation of explainability we introduce KnowEEG; a novel explainable machine learning approach for EEG classification. KnowEEG extracts a comprehensive set of per-electrode features, filters them using statistical tests, and integrates between-electrode connectivity statistics. These features are then input to our modified Random Forest model (Fusion Forest) that balances per electrode statistics with between electrode connectivity features in growing the trees of the forest. By incorporating knowledge from both the generalized time-series and EEG-specific domains, KnowEEG achieves performance comparable to or exceeding state-of-the-art deep learning models across five different classification tasks: emotion detection, mental workload classification, eyes open/closed detection, abnormal EEG classification, and event detection. In addition to high performance, KnowEEG provides inherent explainability through feature importance scores for understandable features. We demonstrate by example on the eyes closed/open classification task that this explainability can be used to discover knowledge about the classes. This discovered knowledge for eyes open/closed classification was proven to be correct by current neuroscience literature. Therefore, the impact of KnowEEG will be significant for domains where EEG explainability is critical such as healthcare.

Nawid Keshtmand, Elena Fillola, Jeffrey Nicholas Clark et al.

Data-driven emulators are increasingly being used to learn and emulate physics-based simulations, reducing computational expense and run time. Here, we present a structured way to improve the quality of these high-dimensional emulated outputs, through the use of prototypes: an approximation of the emulator's output passed as an input, which informs the model and leads to better predictions. We demonstrate our approach to emulate atmospheric dispersion, key for greenhouse gas emissions monitoring, by comparing a baseline model to models trained using prototypes as an additional input. The prototype models achieve better performance, even with few prototypes and even if they are chosen at random, but we show that choosing the prototypes through data-driven methods (k-means) can lead to almost 10\% increased performance in some metrics.

Finn Gueterbock, Raul Santos-Rodriguez, Jeffrey N. Clark

Air pollution is a significant global health risk, contributing to millions of premature deaths annually. Nitrogen dioxide (NO2), a harmful pollutant, disproportionately affects urban areas where monitoring networks are often sparse. We propose a novel method for predicting NO2 concentrations at unmonitored locations using transfer learning with satellite and meteorological data. Leveraging the GraphSAGE framework, our approach integrates autoregression and transfer learning to enhance predictive accuracy in data-scarce regions like Bristol. Pre-trained on data from London, UK, our model achieves a 8.6% reduction in Normalised Root Mean Squared Error (NRMSE) and a 32.6% reduction in Gradient RMSE compared to a baseline model. This work demonstrates the potential of virtual sensors for cost-effective air quality monitoring, contributing to actionable insights for climate and health interventions.

Yujie Dai, Brian Sullivan, Axel Montout et al.

The use of machine learning and AI on electronic health records (EHRs) holds substantial potential for clinical insight. However, this approach faces challenges due to data heterogeneity, sparsity, temporal misalignment, and limited labeled outcomes. In this context, we leverage a linked EHR dataset of approximately one million de-identified individuals from Bristol, North Somerset, and South Gloucestershire, UK, to characterize urinary tract infections (UTIs). We implemented a data pre-processing and curation pipeline that transforms the raw EHR data into a structured format suitable for developing predictive models focused on data fairness, accountability and transparency. Given the limited availability and biases of ground truth UTI outcomes, we introduce a UTI risk estimation framework informed by clinical expertise to estimate UTI risk across individual patient timelines. Pairwise XGBoost models are trained using this framework to differentiate UTI risk categories with explainable AI techniques applied to identify key predictors and support interpretability. Our findings reveal differences in clinical and demographic predictors across risk groups. While this study highlights the potential of AI-driven insights to support UTI clinical decision-making, further investigation of patient sub-strata and extensive validation are needed to ensure robustness and applicability in clinical practice.

Alexander Hepburn, Raul Santos-Rodriguez, Javier Portilla

The Two Alternative Forced Choice (2AFC) paradigm offers advantages over the Mean Opinion Score (MOS) paradigm in psychophysics (PF), such as simplicity and robustness. However, when evaluating perceptual distance models, MOS enables direct correlation between model predictions and PF data. In contrast, 2AFC only allows pairwise comparisons to be converted into a quality ranking similar to MOS when comparisons include shared images. In large datasets, like BAPPS, where image patches and distortions are combined randomly, deriving rankings from 2AFC PF data becomes infeasible, as distorted images included in each comparisons are independent. To address this, instead of relying on MOS correlation, researchers have trained ad-hoc neural networks to reproduce 2AFC PF data based on pairs of model distances - a black-box approach with conceptual and operational limitations. This paper introduces a more robust distance-model evaluation method using a pure probabilistic approach, applying maximum likelihood estimation to a binomial decision model. Our method demonstrates superior simplicity, interpretability, flexibility, and computational efficiency, as shown through evaluations of various visual distance models on two 2AFC PF datasets.

Tashi Namgyal, Peter Flach, Raul Santos-Rodriguez

We describe a proof-of-principle implementation of a system for drawing melodies that abstracts away from a note-level input representation via melodic contours. The aim is to allow users to express their musical intentions without requiring prior knowledge of how notes fit together melodiously. Current approaches to controllable melody generation often require users to choose parameters that are static across a whole sequence, via buttons or sliders. In contrast, our method allows users to quickly specify how parameters should change over time by drawing a contour.

Tashi Namgyal, Alexander Hepburn, Raul Santos-Rodriguez et al.

In this study, we investigate the feasibility of utilizing state-of-the-art image perceptual metrics for evaluating audio signals by representing them as spectrograms. The encouraging outcome of the proposed approach is based on the similarity between the neural mechanisms in the auditory and visual pathways. Furthermore, we customise one of the metrics which has a psychoacoustically plausible architecture to account for the peculiarities of sound signals. We evaluate the effectiveness of our proposed metric and several baseline metrics using a music dataset, with promising results in terms of the correlation between the metrics and the perceived quality of audio as rated by human evaluators.

Rafael Poyiadzi, Daniel Bacaicoa-Barber, Jesus Cid-Sueiro et al.

Many ways of annotating a dataset for machine learning classification tasks that go beyond the usual class labels exist in practice. These are of interest as they can simplify or facilitate the collection of annotations, while not greatly affecting the resulting machine learning model. Many of these fall under the umbrella term of weak labels or annotations. However, it is not always clear how different alternatives are related. In this paper we propose a framework for categorising weak supervision settings with the aim of: (1) helping the dataset owner or annotator navigate through the available options within weak supervision when prescribing an annotation process, and (2) describing existing annotations for a dataset to machine learning practitioners so that we allow them to understand the implications for the learning process. To this end, we identify the key elements that characterise weak supervision and devise a series of dimensions that categorise most of the existing approaches. We show how common settings in the literature fit within the framework and discuss its possible uses in practice.

Telmo Silva Filho, Hao Song, Miquel Perello-Nieto et al.

This paper provides both an introduction to and a detailed overview of the principles and practice of classifier calibration. A well-calibrated classifier correctly quantifies the level of uncertainty or confidence associated with its instance-wise predictions. This is essential for critical applications, optimal decision making, cost-sensitive classification, and for some types of context change. Calibration research has a rich history which predates the birth of machine learning as an academic field by decades. However, a recent increase in the interest on calibration has led to new methods and the extension from binary to the multiclass setting. The space of options and issues to consider is large, and navigating it requires the right set of concepts and tools. We provide both introductory material and up-to-date technical details of the main concepts and methods, including proper scoring rules and other evaluation metrics, visualisation approaches, a comprehensive account of post-hoc calibration methods for binary and multiclass classification, and several advanced topics.

Jonas Schulz, Rafael Poyiadzi, Raul Santos-Rodriguez

Explainability of black-box machine learning models is crucial, in particular when deployed in critical applications such as medicine or autonomous cars. Existing approaches produce explanations for the predictions of models, however, how to assess the quality and reliability of such explanations remains an open question. In this paper we take a step further in order to provide the practitioner with tools to judge the trustworthiness of an explanation. To this end, we produce estimates of the uncertainty of a given explanation by measuring the ordinal consensus amongst a set of diverse bootstrapped surrogate explainers. While we encourage diversity by using ensemble techniques, we propose and analyse metrics to aggregate the information contained within the set of explainers through a rating scheme. We empirically illustrate the properties of this approach through experiments on state-of-the-art Convolutional Neural Network ensembles. Furthermore, through tailored visualisations, we show specific examples of situations where uncertainty estimates offer concrete actionable insights to the user beyond those arising from standard surrogate explainers.

Taku Yamagata, Ryan McConville, Raul Santos-Rodriguez

A promising approach to improve the robustness and exploration in Reinforcement Learning is collecting human feedback and that way incorporating prior knowledge of the target environment. It is, however, often too expensive to obtain enough feedback of good quality. To mitigate the issue, we aim to rely on a group of multiple experts (and non-experts) with different skill levels to generate enough feedback. Such feedback can therefore be inconsistent and infrequent. In this paper, we build upon prior work -- Advise, a Bayesian approach attempting to maximise the information gained from human feedback -- extending the algorithm to accept feedback from this larger group of humans, the trainers, while also estimating each trainer's reliability. We show how aggregating feedback from multiple trainers improves the total feedback's accuracy and make the collection process easier in two ways. Firstly, this approach addresses the case of some of the trainers being adversarial. Secondly, having access to the information about each trainer reliability provides a second layer of robustness and offers valuable information for people managing the whole system to improve the overall trust in the system. It offers an actionable tool for improving the feedback collection process or modifying the reward function design if needed. We empirically show that our approach can accurately learn the reliability of each trainer correctly and use it to maximise the information gained from the multiple trainers' feedback, even if some of the sources are adversarial.

Rafael Poyiadzi, Xavier Renard, Thibault Laugel et al.

This paper analyses the fundamental ingredients behind surrogate explanations to provide a better understanding of their inner workings. We start our exposition by considering global surrogates, describing the trade-off between complexity of the surrogate and fidelity to the black-box being modelled. We show that transitioning from global to local - reducing coverage - allows for more favourable conditions on the Pareto frontier of fidelity-complexity of a surrogate. We discuss the interplay between complexity, fidelity and coverage, and consider how different user needs can lead to problem formulations where these are either constraints or penalties. We also present experiments that demonstrate how the local surrogate interpretability procedure can be made interactive and lead to better explanations.

Rafael Poyiadzi, Xavier Renard, Thibault Laugel et al.

Local surrogate approaches for explaining machine learning model predictions have appealing properties, such as being model-agnostic and flexible in their modelling. Several methods exist that fit this description and share this goal. However, despite their shared overall procedure, they set out different objectives, extract different information from the black-box, and consequently produce diverse explanations, that are -- in general -- incomparable. In this work we review the similarities and differences amongst multiple methods, with a particular focus on what information they extract from the model, as this has large impact on the output: the explanation. We discuss the implications of the lack of agreement, and clarity, amongst the methods' objectives on the research and practice of explainability.

Alexander Hepburn, Valero Laparra, Raul Santos-Rodriguez et al.

It has been demonstrated many times that the behavior of the human visual system is connected to the statistics of natural images. Since machine learning relies on the statistics of training data as well, the above connection has interesting implications when using perceptual distances (which mimic the behavior of the human visual system) as a loss function. In this paper, we aim to unravel the non-trivial relationships between the probability distribution of the data, perceptual distances, and unsupervised machine learning. To this end, we show that perceptual sensitivity is correlated with the probability of an image in its close neighborhood. We also explore the relation between distances induced by autoencoders and the probability distribution of the training data, as well as how these induced distances are correlated with human perception. Finally, we find perceptual distances do not always lead to noticeable gains in performance over Euclidean distance in common image processing tasks, except when data is scarce and the perceptual distance provides regularization. We propose this may be due to a \emph{double-counting} effect of the image statistics, once in the perceptual distance and once in the training procedure.

Hok-Shing Lau, Ryan McConville, Mohammud J. Bocus et al.

Traditional approaches to activity recognition involve the use of wearable sensors or cameras in order to recognise human activities. In this work, we extract fine-grained physical layer information from WiFi devices for the purpose of passive activity recognition in indoor environments. While such data is ubiquitous, few approaches are designed to utilise large amounts of unlabelled WiFi data. We propose the use of self-supervised contrastive learning to improve activity recognition performance when using multiple views of the transmitted WiFi signal captured by different synchronised receivers. We conduct experiments where the transmitters and receivers are arranged in different physical layouts so as to cover both Line-of-Sight (LoS) and non LoS (NLoS) conditions. We compare the proposed contrastive learning system with non-contrastive systems and observe a 17.7% increase in macro averaged F1 score on the task of WiFi based activity recognition, as well as significant improvements in one- and few-shot learning scenarios.

Xiaoyang Wang, Jonathan D Thomas, Robert J Piechocki et al.

Open Radio Access Network (ORAN) is being developed with an aim to democratise access and lower the cost of future mobile data networks, supporting network services with various QoS requirements, such as massive IoT and URLLC. In ORAN, network functionality is dis-aggregated into remote units (RUs), distributed units (DUs) and central units (CUs), which allows flexible software on Commercial-Off-The-Shelf (COTS) deployments. Furthermore, the mapping of variable RU requirements to local mobile edge computing centres for future centralized processing would significantly reduce the power consumption in cellular networks. In this paper, we study the RU-DU resource assignment problem in an ORAN system, modelled as a 2D bin packing problem. A deep reinforcement learning-based self-play approach is proposed to achieve efficient RU-DU resource management, with AlphaGo Zero inspired neural Monte-Carlo Tree Search (MCTS). Experiments on representative 2D bin packing environment and real sites data show that the self-play learning strategy achieves intelligent RU-DU resource assignment for different network conditions.

Rafael Poyiadzi, Weisong Yang, Niall Twomey et al.

In this paper we provide machine learning practitioners with tools to answer the question: is there class-conditional noise in my labels? In particular, we present hypothesis tests to check whether a given dataset of instance-label pairs has been corrupted with class-conditional label noise, as opposed to uniform label noise, with the former biasing learning, while the latter -- under mild conditions -- does not. The outcome of these tests can then be used in conjunction with other information to assess further steps. While previous works explore the direct estimation of the noise rates, this is known to be hard in practice and does not offer a real understanding of how trustworthy the estimates are. These methods typically require anchor points -- examples whose true posterior is either 0 or 1. Differently, in this paper we assume we have access to a set of anchor points whose true posterior is approximately 1/2. The proposed hypothesis tests are built upon the asymptotic properties of Maximum Likelihood Estimators for Logistic Regression models. We establish the main properties of the tests, including a theoretical and empirical analysis of the dependence of the power on the test on the training sample size, the number of anchor points, the difference of the noise rates and the use of relaxed anchors.

Alexander Hepburn, Raul Santos-Rodriguez

Explaining the decisions of models is becoming pervasive in the image processing domain, whether it is by using post-hoc methods or by creating inherently interpretable models. While the widespread use of surrogate explainers is a welcome addition to inspect and understand black-box models, assessing the robustness and reliability of the explanations is key for their success. Additionally, whilst existing work in the explainability field proposes various strategies to address this problem, the challenges of working with data in the wild is often overlooked. For instance, in image classification, distortions to images can not only affect the predictions assigned by the model, but also the explanation. Given a clean and a distorted version of an image, even if the prediction probabilities are similar, the explanation may still be different. In this paper we propose a methodology to evaluate the effect of distortions in explanations by embedding perceptual distances that tailor the neighbourhoods used to training surrogate explainers. We also show that by operating in this way, we can make the explanations more robust to distortions. We generate explanations for images in the Imagenet-C dataset and demonstrate how using a perceptual distances in the surrogate explainer creates more coherent explanations for the distorted and reference images.

Rafael Poyiadzi, Weisong Yang, Yoav Ben-Shlomo et al.

There is a pressing need to automatically understand the state and progression of chronic neurological diseases such as dementia. The emergence of state-of-the-art sensing platforms offers unprecedented opportunities for indirect and automatic evaluation of disease state through the lens of behavioural monitoring. This paper specifically seeks to characterise behavioural signatures of mild cognitive impairment (MCI) and Alzheimer's disease (AD) in the \textit{early} stages of the disease. We introduce bespoke behavioural models and analyses of key symptoms and deploy these on a novel dataset of longitudinal sensor data from persons with MCI and AD. We present preliminary findings that show the relationship between levels of sleep quality and wandering can be subtly different between patients in the early stages of dementia and healthy cohabiting controls.

Kacper Sokol, Alexander Hepburn, Raul Santos-Rodriguez et al.

Surrogate explainers of black-box machine learning predictions are of paramount importance in the field of eXplainable Artificial Intelligence since they can be applied to any type of data (images, text and tabular), are model-agnostic and are post-hoc (i.e., can be retrofitted). The Local Interpretable Model-agnostic Explanations (LIME) algorithm is often mistakenly unified with a more general framework of surrogate explainers, which may lead to a belief that it is the solution to surrogate explainability. In this paper we empower the community to "build LIME yourself" (bLIMEy) by proposing a principled algorithmic framework for building custom local surrogate explainers of black-box model predictions, including LIME itself. To this end, we demonstrate how to decompose the surrogate explainers family into algorithmically independent and interoperable modules and discuss the influence of these component choices on the functional capabilities of the resulting explainer, using the example of LIME.