Large Language Models (LLMs) have been found to struggle with systematic reasoning. Even on tasks where they appear to perform well, their performance often depends on shortcuts, rather than on genuine reasoning abilities, leading them to collapse on out-of-distribution examples. Post-training strategies based on reinforcement learning and chain-of-thought prompting have recently been hailed as a step change. However, little is still known about the potential of the resulting ``Large Reasoning Models'' (LRMs) beyond problem solving in mathematics and programming, where finding genuine out-of-distribution problems can be difficult. In this paper, we focus on tasks that require systematic reasoning about relational compositions, especially for qualitative spatial and temporal reasoning. These tasks allow us to control the difficulty of problem instances, and measure in a precise way to what extent models can generalise. We find that that the considered LLMs and LRMs overall perform poorly overall, albeit better than random chance.

Decentralized federated learning enables learning of data representations from multiple sources without compromising the privacy of the clients. In applications like medical image segmentation, where obtaining a large annotated dataset from a single source is a distressing problem, federated self-supervised learning can provide some solace. In this work, we push the limits further by exploring a federated self-supervised one-shot segmentation task representing a more data-scarce scenario. We adopt a pre-existing self-supervised few-shot segmentation framework CoWPro and adapt it to the federated learning scenario. To the best of our knowledge, this work is the first to attempt a self-supervised few-shot segmentation task in the federated learning domain. Moreover, we consider the clients to be constituted of data from different modalities and imaging techniques like MR or CT, which makes the problem even harder. Additionally, we reinforce and improve the baseline CoWPro method using a fused dice loss which shows considerable improvement in performance over the baseline CoWPro. Finally, we evaluate this novel framework on a completely unseen held-out part of the local client dataset. We observe that the proposed framework can achieve performance at par or better than the FedAvg version of the CoWPro framework on the held-out validation dataset.

Multilingual Natural Language Generation (NLG) is challenging due to the lack of training data for low-resource languages. However, some low-resource languages have up to tens of millions of speakers globally, making it important to improve NLG tools for them. Table-to-Text NLG is an excellent measure of models' reasoning abilities but is very challenging in the multilingual setting. System outputs are often not attributable, or faithful, to the data in the source table. Intermediate planning techniques like Question-Answer (QA) blueprints have been shown to improve attributability on summarisation tasks. This work explores whether QA blueprints make multilingual Table-to-Text outputs more attributable to the input tables. This paper extends the challenging multilingual Table-to-Text dataset, TaTA, which includes African languages, with QA blueprints. Sequence-to-sequence language models are then finetuned on this dataset, with and without blueprints. Results show that QA blueprints improve performance for models finetuned and evaluated only on English examples, but do not demonstrate gains in the multilingual setting. This is due to inaccuracies in machine translating the blueprints from English into target languages when generating the training data, and models failing to rely closely on the blueprints they generate. An in-depth analysis is conducted on why this is challenging.

Large language Models (LLMs) have demonstrated remarkable skills across various domains. Understanding the mechanisms behind their abilities and implementing controls over them is becoming increasingly important for developing better models. In this paper, we focus on skill unlearning in LLMs, specifically unlearning a particular skill while retaining their overall capabilities. We introduce two lightweight, training-free machine skill unlearning techniques for LLMs. First, we observe that the pre-activation distribution of neurons in each Feed-Forward Layer (FFL) differs when the model demonstrates different skills. Additionally, we find that queries triggering the same skill cluster within the FFL key space and can be separated from other queries using a hypercube. Based on these observations, we propose two lightweight, training-free skill unlearning methods via \textit{intervention} and \textit{abstention} respectively: \texttt{Neuron Adjust} and \texttt{Key Space Detection}. We evaluate our methods on unlearning math-solving, Python-coding, and comprehension skills across seven different languages. The results demonstrate their strong unlearning capabilities for the designated skills. Specifically, \texttt{Key Space Detection} achieves over 80\% relative performance drop on the forgetting skill and less than 10\% relative performance drop on other skills and the model's general knowledge (MMLU) for most unlearning tasks. Our code is available at https://github.com/Trustworthy-ML-Lab/effective_skill_unlearning

The rapid growth of artificial intelligence (AI) technologies has changed decision-making in many fields. But, it has also raised major privacy and ethical concerns. However, many AI incidents taxonomies and guidelines for academia, industry, and government lack grounding in real-world incidents. We analyzed 202 real-world AI privacy and ethical incidents. This produced a taxonomy that classifies incident types across AI lifecycle stages. It accounts for contextual factors such as causes, responsible entities, disclosure sources, and impacts. Our findings show insufficient incident reporting from AI developers and users. Many incidents are caused by poor organizational decisions and legal non-compliance. Only a few legal actions and corrective measures exist, while risk-mitigation efforts are limited. Our taxonomy contributes a structured approach in reporting of future AI incidents. Our findings demonstrate that current AI governance frameworks are inadequate. We urgently need child-specific protections and AI policies on social media. They must moderate and reduce the spread of harmful AI-generated content. Our research provides insights for policymakers and practitioners, which lets them design ethical AI. It also support AI incident detection and risk management. Finally, it guides AI policy development. Improved policies will protect people from harmful AI applications and support innovation in AI systems.

Fairness studies of algorithmic decision-making systems often simplify complex decision processes, such as bail or loan approvals, into binary classification tasks. However, these approaches overlook that such decisions are not inherently binary (e.g., approve or not approve bail or loan); they also involve non-binary treatment decisions (e.g., bail conditions or loan terms) that can influence the downstream outcomes (e.g., loan repayment or reoffending). In this paper, we argue that non-binary treatment decisions are integral to the decision process and controlled by decision-makers and, therefore, should be central to fairness analyses in algorithmic decision-making. We propose a causal framework that extends fairness analyses and explicitly distinguishes between decision-subjects' covariates and the treatment decisions. This specification allows decision-makers to use our framework to (i) measure treatment disparity and its downstream effects in historical data and, using counterfactual reasoning, (ii) mitigate the impact of past unfair treatment decisions when automating decision-making. We use our framework to empirically analyze four widely used loan approval datasets to reveal potential disparity in non-binary treatment decisions and their discriminatory impact on outcomes, highlighting the need to incorporate treatment decisions in fairness assessments. Moreover, by intervening in treatment decisions, we show that our framework effectively mitigates treatment discrimination from historical data to ensure fair risk score estimation and (non-binary) decision-making processes that benefit all stakeholders.

The integration of multi-omics data presents a major challenge in precision medicine, requiring advanced computational methods for accurate disease classification and biological interpretation. This study introduces the Multi-Omics Graph Kolmogorov-Arnold Network (MOGKAN), a deep learning model that integrates messenger RNA, micro RNA sequences, and DNA methylation data with Protein-Protein Interaction (PPI) networks for accurate and interpretable cancer classification across 31 cancer types. MOGKAN employs a hybrid approach combining differential expression with DESeq2, Linear Models for Microarray (LIMMA), and Least Absolute Shrinkage and Selection Operator (LASSO) regression to reduce multi-omics data dimensionality while preserving relevant biological features. The model architecture is based on the Kolmogorov-Arnold theorem principle, using trainable univariate functions to enhance interpretability and feature analysis. MOGKAN achieves classification accuracy of 96.28 percent and demonstrates low experimental variability with a standard deviation that is reduced by 1.58 to 7.30 percents compared to Convolutional Neural Networks (CNNs) and Graph Neural Networks (GNNs). The biomarkers identified by MOGKAN have been validated as cancer-related markers through Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis. The proposed model presents an ability to uncover molecular oncogenesis mechanisms by detecting phosphoinositide-binding substances and regulating sphingolipid cellular processes. By integrating multi-omics data with graph-based deep learning, our proposed approach demonstrates superior predictive performance and interpretability that has the potential to enhance the translation of complex multi-omics data into clinically actionable cancer diagnostics.

Congenital heart disease (CHD) is a critical condition that demands early detection, particularly in infancy and childhood. This study presents a deep learning model designed to detect CHD using phonocardiogram (PCG) signals, with a focus on its application in global health. We evaluated our model on several datasets, including the primary dataset from Bangladesh, achieving a high accuracy of 94.1%, sensitivity of 92.7%, specificity of 96.3%. The model also demonstrated robust performance on the public PhysioNet Challenge 2022 and 2016 datasets, underscoring its generalizability to diverse populations and data sources. We assessed the performance of the algorithm for single and multiple auscultation sites on the chest, demonstrating that the model maintains over 85% accuracy even when using a single location. Furthermore, our algorithm was able to achieve an accuracy of 80% on low-quality recordings, which cardiologists deemed non-diagnostic. This research suggests that an AI- driven digital stethoscope could serve as a cost-effective screening tool for CHD in resource-limited settings, enhancing clinical decision support and ultimately improving patient outcomes.

Recent shifts in the space of large language model (LLM) research have shown an increasing focus on novel architectures to compete with prototypical Transformer-based models that have long dominated this space. Linear recurrent models have proven to be a viable competitor due to their computational efficiency. However, such models still demonstrate a sizable gap compared to Transformers in terms of in-context learning among other tasks that require recalling information from a context. In this work, we introduce __Resona__, a simple and scalable framework for augmenting linear recurrent models with retrieval. __Resona__~augments models with the ability to integrate retrieved information from the provided input context, enabling tailored behavior to diverse task requirements. Experiments on a variety of linear recurrent models demonstrate that __Resona__-augmented models observe significant performance gains on a variety of synthetic as well as real-world natural language tasks, highlighting its ability to act as a general purpose method to improve the in-context learning and language modeling abilities of linear recurrent LLMs.

Tweets provides valuable semantic context for earth observation tasks and serves as a complementary modality to remote sensing imagery. In building function classification (BFC), tweets are often collected using geographic heuristics and labeled via external databases, an inherently weakly supervised process that introduces both label noise and sentence level feature noise (e.g., irrelevant or uninformative tweets). While label noise has been widely studied, the impact of sentence level feature noise remains underexplored, largely due to the lack of clean benchmark datasets for controlled analysis. In this work, we propose a method for generating a synthetic oracle dataset using LLM, designed to contain only tweets that are both correctly labeled and semantically relevant to their associated buildings. This oracle dataset enables systematic investigation of noise impacts that are otherwise difficult to isolate in real-world data. To assess its utility, we compare model performance using Naive Bayes and mBERT classifiers under three configurations: real vs. synthetic training data, and cross-domain generalization. Results show that noise in real tweets significantly degrades the contextual learning capacity of mBERT, reducing its performance to that of a simple keyword-based model. In contrast, the clean synthetic dataset allows mBERT to learn effectively, outperforming Naive Bayes Bayes by a large margin. These findings highlight that addressing feature noise is more critical than model complexity in this task. Our synthetic dataset offers a novel experimental environment for future noise injection studies and is publicly available on GitHub.