Whiletheoretical results havebeen obtained for their heavy-tailedness, the full characterization of the prior predictive distribution (i.e. its density, CDF andmoments), remained unknownpriortothiswork.

Recent works on Bayesian neural networks (BNNs) have highlighted the need to better understand the implications of using Gaussian priors in combination with the compositional structure of the network architecture. Similar in spirit to the kind of analysis that has been developed to devise better initialization schemes for neural networks (cf. He-or Xavier initialization), we derive a precise characterization of the prior predictive distribution of finite-width ReLU networks with Gaussian weights. While theoretical results have been obtained for their heavy-tailedness, the full characterization of the prior predictive distribution (i.e. its density, CDF and moments), remained unknown prior to this work. Our analysis, based on the Meijer-G function, allows us to quantify the influence of architectural choices such as the width or depth of the network on the resulting shape of the prior predictive distribution. We also formally connect our results to previous work in the infinite width setting, demonstrating that the moments of the distribution converge to those of a normal log-normal mixture in the infinite depth limit. Finally, our results provide valuable guidance on prior design: for instance, controlling the predictive variance with depth-and width-informed priors on the weights of the network.

Photoplethysmography (PPG) signals encode information about relative changes in blood volume that can be used to assess various aspects of cardiac health non-invasively, e.g.\ to detect atrial fibrillation (AF) or predict blood pressure (BP). Deep networks are well-equipped to handle the large quantities of data acquired from wearable measurement devices. However, they lack interpretability and are prone to overfitting, leaving considerable risk for poor performance on unseen data and misdiagnosis. Here, we describe the use of two scalable uncertainty quantification techniques: Monte Carlo Dropout and the recently proposed Improved Variational Online Newton. These techniques are used to assess the trustworthiness of models trained to perform AF classification and BP regression from raw PPG time series. We find that the choice of hyperparameters has a considerable effect on the predictive performance of the models and on the quality and composition of predicted uncertainties. E.g. the stochasticity of the model parameter sampling determines the proportion of the total uncertainty that is aleatoric, and has varying effects on predictive performance and calibration quality dependent on the chosen uncertainty quantification technique and the chosen expression of uncertainty. We find significant discrepancy in the quality of uncertainties over the predicted classes, emphasising the need for a thorough evaluation protocol that assesses local and adaptive calibration. This work suggests that the choice of hyperparameters must be carefully tuned to balance predictive performance and calibration quality, and that the optimal parameterisation may vary depending on the chosen expression of uncertainty.

Multimodal models integrating speech and vision hold significant potential for advancing human-computer interaction, particularly in Speech-Based Visual Question Answering (SBVQA) where spoken questions about images require direct audio-visual understanding. Existing approaches predominantly focus on text-visual integration, leaving speech-visual modality gaps underexplored due to their inherent heterogeneity. To this end, we introduce SViQA, a unified speech-vision model that directly processes spoken questions without text transcription. Building upon the LLaVA architecture, our framework bridges auditory and visual modalities through two key innovations: (1) end-to-end speech feature extraction eliminating intermediate text conversion, and (2) cross-modal alignment optimization enabling effective fusion of speech signals with visual content. Extensive experimental results on the SBVQA benchmark demonstrate the proposed SViQA's state-of-the-art performance, achieving 75.62% accuracy, and competitive multimodal generalization. Leveraging speech-text mixed input boosts performance to 78.85%, a 3.23% improvement over pure speech input, highlighting SViQA's enhanced robustness and effective cross-modal attention alignment.

Child bicyclists (14 years and younger) are among the most vulnerable road users, often experiencing severe injuries or fatalities in crashes. This study analyzed 2,394 child bicyclist crashes in Texas from 2017 to 2022 using two deep tabular learning models (ARM-Net and MambaNet). To address the issue of data imbalance, the SMOTEENN technique was applied, resulting in balanced datasets that facilitated accurate crash severity predictions across three categories: Fatal/Severe (KA), Moderate/Minor (BC), and No Injury (O). The findings revealed that MambaNet outperformed ARM-Net, achieving higher precision, recall, F1-scores, and accuracy, particularly in the KA and O categories. Both models highlighted challenges in distinguishing BC crashes due to overlapping characteristics. These insights underscored the value of advanced tabular deep learning methods and balanced datasets in understanding crash severity. While limitations such as reliance on categorical data exist, future research could explore continuous variables and real-time behavioral data to enhance predictive modeling and crash mitigation strategies.

Young motorcyclists, particularly those aged 15 to 24 years old, face a heightened risk of severe crashes due to factors such as speeding, traffic violations, and helmet usage. This study aims to identify key factors influencing crash severity by analyzing 10,726 young motorcyclist crashes in Texas from 2017 to 2022. Two advanced tabular deep learning models, ARMNet and MambaNet, were employed, using an advanced resampling technique to address class imbalance. The models were trained to classify crashes into three severity levels, Fatal or Severe, Moderate or Minor, and No Injury. ARMNet achieved an accuracy of 87 percent, outperforming 86 percent of Mambanet, with both models excelling in predicting severe and no injury crashes while facing challenges in moderate crash classification. Key findings highlight the significant influence of demographic, environmental, and behavioral factors on crash outcomes. The study underscores the need for targeted interventions, including stricter helmet enforcement and educational programs customized to young motorcyclists. These insights provide valuable guidance for policymakers in developing evidence-based strategies to enhance motorcyclist safety and reduce crash severity.

Recent advancements in large language models (LLMs) have led to their widespread adoption and large-scale deployment across various domains. However, their environmental impact, particularly during inference, has become a growing concern due to their substantial energy consumption and carbon footprint. Existing research has focused on inference computation alone, overlooking the analysis and optimization of carbon footprint in network-aided LLM service systems. To address this gap, we propose AOLO, a framework for analysis and optimization for low-carbon oriented wireless LLM services. AOLO introduces a comprehensive carbon footprint model that quantifies greenhouse gas emissions across the entire LLM service chain, including computational inference and wireless communication. Furthermore, we formulate an optimization problem aimed at minimizing the overall carbon footprint, which is solved through joint optimization of inference outputs and transmit power under quality-of-experience and system performance constraints. To achieve this joint optimization, we leverage the energy efficiency of spiking neural networks (SNNs) by adopting SNN as the actor network and propose a low-carbon-oriented optimization algorithm, i.e., SNN-based deep reinforcement learning (SDRL). Comprehensive simulations demonstrate that SDRL algorithm significantly reduces overall carbon footprint, achieving an 18.77% reduction compared to the benchmark soft actor-critic, highlighting its potential for enabling more sustainable LLM inference services.

In recent years, numerous neuroscientific studies have shown that human emotions are closely linked to specific brain regions, with these regions exhibiting variability across individuals and emotional states. To fully leverage these neural patterns, we propose an Adaptive Progressive Attention Graph Neural Network (APAGNN), which dynamically captures the spatial relationships among brain regions during emotional processing. The APAGNN employs three specialized experts that progressively analyze brain topology. The first expert captures global brain patterns, the second focuses on region-specific features, and the third examines emotion-related channels. This hierarchical approach enables increasingly refined analysis of neural activity. Additionally, a weight generator integrates the outputs of all three experts, balancing their contributions to produce the final predictive label. Extensive experiments on three publicly available datasets (SEED, SEED-IV and MPED) demonstrate that the proposed method enhances EEG emotion recognition performance, achieving superior results compared to baseline methods.

This paper proposes small and efficient machine learning models (TinyML) for resource-constrained edge devices, specifically for on-device indoor localisation. Typical approaches for indoor localisation rely on centralised remote processing of data transmitted from lower powered devices such as wearables. However, there are several benefits for moving this to the edge device itself, including increased battery life, enhanced privacy, reduced latency and lowered operational costs, all of which are key for common applications such as health monitoring. The work focuses on model compression techniques, including quantization and knowledge distillation, to significantly reduce the model size while maintaining high predictive performance. We base our work on a large state-of-the-art transformer-based model and seek to deploy it within low-power MCUs. We also propose a state-space-based architecture using Mamba as a more compact alternative to the transformer. Our results show that the quantized transformer model performs well within a 64 KB RAM constraint, achieving an effective balance between model size and localisation precision. Additionally, the compact Mamba model has strong performance under even tighter constraints, such as a 32 KB of RAM, without the need for model compression, making it a viable option for more resource-limited environments. We demonstrate that, through our framework, it is feasible to deploy advanced indoor localisation models onto low-power MCUs with restricted memory limitations. The application of these TinyML models in healthcare has the potential to revolutionize patient monitoring by providing accurate, real-time location data while minimizing power consumption, increasing data privacy, improving latency and reducing infrastructure costs.

With the growing percentage of elderly people and care home admissions, there is an urgent need for the development of fall detection and fall prevention technologies. This work presents, for the first time, the use of machine learning techniques to recognize postural movements exclusively from Photoplethysmography (PPG) data. To achieve this goal, a device was developed for reading the PPG signal, segmenting the PPG signals into individual pulses, extracting pulse morphology and homeostatic characteristic features, and evaluating different ML algorithms. Investigations into different postural movements (stationary, sitting to standing, and lying to standing) were performed by 11 participants. The results of these investigations provided insight into the differences in homeostasis after the movements in the PPG signal. Various machine learning approaches were used for classification, and the Artificial Neural Network (ANN) was found to be the best classifier, with a testing accuracy of 85.2\% and an F1 score of 78\% from experimental results.