Proceedings of the UK AI Conference 2024 Held in The Exchange, University of Birmingham, Birmingham, United Kingdom on 22 November 2024 Published as Volume 295 by the Proceedings of Machine Learning Research on 05 August 2025. Volume Edited by: Alistair Benford Christian Cabrera Sarah Kiden Arianna Salili-James Vincent Zakka Gbouna Series Editors: Neil D. Lawrence https://proceedings.mlr.press/v295/ Tue, 12 Aug 2025 13:39:39 +0000 Tue, 12 Aug 2025 13:39:39 +0000 Jekyll v3.10.0 This paper conducted a comparative evaluation two approaches for predicting word complexity using contextual sentence information, a challenge that traditional methods often struggle to address. Two distinct methods were explored in this work. The first approach combines XLNet word embeddings with a Random Forest classifier to processes both sentence and word embeddings to predict complexity levels. The second approach employs a dual Bidirectional Encoder Representations from Transformers (BERT) model, consisting of two separate models: one for sentence-level complexity and another for word-level complexity, with their predictions combined for more context-sensitive result. A diverse dataset covering the domains of religion, biomedical, and parliamentary texts was used, as it is pre-categorised into five complexity levels (Very-easy, Easy, Medium, Hard, Very-hard). To ensure balanced class representation, data augmentation techniques were applied. Evaluation metrics revealed that the XLNet-based model has performed slightly superior to dual-BERT method, achieving macro-average F1-score of 0.79, excelling particularly at identifying highly complex words (F1-score = 0.95). In comparison, dual-BERT achieved a macro-average F1-score equal to 0.78. Tue, 05 Aug 2025 00:00:00 +0000 https://proceedings.mlr.press/v295/uzzam25a.html https://proceedings.mlr.press/v295/uzzam25a.html Accurately estimating uncertainty in predictive models is crucial for a wide range of applications, from decision-making in landuse modelling to robust forecasting in finance and autonomous systems. Gaussian processes (GPs) offer a solid framework for uncertainty quantification but often struggle with scalability and flexibility when applied to large, high-dimensional datasets. Deep Gaussian processes (DGPs) are a powerful extension of GPs that allow for multi-layer generalisation of GPs, enabling more flexible and expressive modelling of complex data. As the complexity of the model increases, so does the computational cost, which makes it difficult to scale DGP to large-dimensional data. Although variational inference has been used with large datasets, it often produces an overconfident uncertainty estimate because it does not effectively utilise input-dependent function uncertainty. This paper introduces an approach for enhancing uncertainty estimation using the predictive log-likelihood (PLL) objective with DGP model to address these limitations. This relies on a parametric GP regression model designed for a family of predictive distributions and incorporate a modified objective function to restore a full symmetry between various contributions to predictive variance. We evaluate the performance of our methods on several benchmark regressions and large-scale environmental datasets. The results show that the model provides more reliable uncertainty estimates, particularly in regions of sparse data, making them efficient for real-world applications. Tue, 05 Aug 2025 00:00:00 +0000 https://proceedings.mlr.press/v295/oyebamiji25a.html https://proceedings.mlr.press/v295/oyebamiji25a.html We explore artist-led research as a method to complement technical AI methodologies. We present a case study called Cat Royale in which artists created a robot to play with cats. We show how the artist-led development of this system involved extensive improvisation to create a socio-technical AI system that ultimately delivered a corpus of video data of cats interacting with robots. We introduce a machine learning tool that enables diverse stakeholders to explore this corpus. We reflect on the distinctive characteristics of artist-led AI research, the potential benefits to AI, and the tensions involved. Tue, 05 Aug 2025 00:00:00 +0000 https://proceedings.mlr.press/v295/cao25a.html https://proceedings.mlr.press/v295/cao25a.html Artificial Intelligence (AI), a rapidly evolving technology with far-reaching implications, has become a widely discussed topic on social media platforms. This study conducted a comprehensive sentiment analysis of posts discussing AI topics on Twitter, YouTube and Reddit from 2017 to 2024 to evaluate public perceptions and attitudes toward AI. A total of 133,004 social media posts were analysed using a fine-tuned RoBERTa model for sentiment classification, alongside Latent Dirichlet Allocation (LDA), n-gram, and word co-occurrence mapping for topic modelling. The analysis revealed that 46.09% of the posts express negative sentiments about AI, followed by 39.29% neutral and 14.62% positive sentiments. LDA uncovered 20 key topics, including AI ethics, job impacts, and philosophical implications. Temporal analysis revealed a significant surge in AI-related discourse from 2022 onward, with evolving sentiment patterns. These findings suggest a complex landscape of public AI perception, reflecting persistent concerns about societal impacts alongside increasing interest in AI’s technical aspects. Tue, 05 Aug 2025 00:00:00 +0000 https://proceedings.mlr.press/v295/appiah25a.html https://proceedings.mlr.press/v295/appiah25a.html The “chicken-and-egg problem” in Electric Vehicle (EV) charging reflects the interdependence between sufficient infrastructure and the demand needed to justify it, a challenge heightened by the UK’s 2030 ban on new combustion engine vehicles. To address this, we propose a joint optimisation model that determines the optimal number of charging points and pricing at each station, while accounting for traffic patterns. From a policy perspective, our model seeks to maximise public benefit by reducing EV users’ social costs, travel and queuing time, and charging fees, while ensuring station operator profitability. We model driver decisions as two interconnected congestion games, one on roads and one at charging stations (CS), and solve for stable outcomes using Nash Equilibrium (NE) strategies. To ensure tractability, we develop an efficient approximation algorithm for the Mixed-Integer Nonlinear Program (MINLP) and introduce a generalisation technique that targets charger placement at high-impact locations, enhancing scalability to larger Transportation Networks (TN). Applied to a benchmark case, the model reduces overall social cost by at least 14% compared to methods that optimise placement or pricing alone. This study tackles an AI challenge in modelling infrastructure with multi-agent behaviour, using game theory and optimisation to simulate interactions and enable learning-based approaches in transportation systems. Tue, 05 Aug 2025 00:00:00 +0000 https://proceedings.mlr.press/v295/aminikalibar25a.html https://proceedings.mlr.press/v295/aminikalibar25a.html