Research Papers

Driven by the rapid progress in vision-language models (VLMs), the responsible behavior of large-scale multimodal models has become a prominent research area, particularly focusing on hallucination detection and factuality checking. In this paper, we present the solution for the two tracks of Responsible AI challenge. Inspirations from the general domain demonstrate that a smaller distilled VLM can often outperform a larger VLM that is directly tuned on downstream tasks, while achieving higher efficiency. We thus jointly tackle two tasks from the perspective of knowledge distillation and propose a progressive hybrid knowledge distillation framework termed HKD4VLM. Specifically, the overall framework can be decomposed into Pyramid-like Progressive Online Distillation and Ternary-Coupled Refinement Distillation, hierarchically moving from coarse-grained knowledge alignment to fine-grained refinement. Besides, we further introduce the mapping shift-enhanced inference and diverse augmentation strategies to enhance model performance and robustness. Extensive experimental results demonstrate the effectiveness of our HKD4VLM. Ablation studies provide insights into the critical design choices driving performance gains.

The recent success of generative AI highlights the crucial role of high-quality human feedback in building trustworthy AI systems. However, the increasing use of large language models (LLMs) by crowdsourcing workers poses a significant challenge: datasets intended to reflect human input may be compromised by LLM-generated responses. Existing LLM detection approaches often rely on high-dimensional training data such as text, making them unsuitable for annotation tasks like multiple-choice labeling. In this work, we investigate the potential of peer prediction -- a mechanism that evaluates the information within workers' responses without using ground truth -- to mitigate LLM-assisted cheating in crowdsourcing with a focus on annotation tasks. Our approach quantifies the correlations between worker answers while conditioning on (a subset of) LLM-generated labels available to the requester. Building on prior research, we propose a training-free scoring mechanism with theoretical guarantees under a crowdsourcing model that accounts for LLM collusion. We establish conditions under which our method is effective and empirically demonstrate its robustness in detecting low-effort cheating on real-world crowdsourcing datasets.

A common practice of ML systems development concerns the training of the same model under different data sets, and the use of the same (training and test) sets for different learning models. The first case is a desirable practice for identifying high quality and unbiased training conditions. The latter case coincides with the search for optimal models under a common dataset for training. These differently obtained systems have been considered akin to copies. In the quest for responsible AI, a legitimate but hardly investigated question is how to verify that trustworthiness is preserved by copies. In this paper we introduce a calculus to model and verify probabilistic complex queries over data and define four distinct notions: Justifiably, Equally, Weakly and Almost Trustworthy which can be checked analysing the (partial) behaviour of the copy with respect to its original. We provide a study of the relations between these notions of trustworthiness, and how they compose with each other and under logical operations. The aim is to offer a computational tool to check the trustworthiness of possibly complex systems copied from an original whose behavour is known.

Automatic reading aloud evaluation can provide valuable support to teachers by enabling more efficient scoring of reading exercises. However, research on reading evaluation systems and applications remains limited. We present a novel multimodal approach that leverages audio and knowledge from text resources. In particular, we explored the potential of using Whisper and instruction-tuned large language models (LLMs) with prompts to improve transcriptions for child speech recognition, as well as their effectiveness in downstream reading mistake detection. Our results demonstrate the effectiveness of prompting Whisper and prompting LLM, compared to the baseline Whisper model without prompting. The best performing system achieved state-of-the-art recognition performance in Dutch child read speech, with a word error rate (WER) of 5.1%, improving the baseline WER of 9.4%. Furthermore, it significantly improved reading mistake detection, increasing the F1 score from 0.39 to 0.73.

XAI-Units: Benchmarking Explainability Methods with Unit Tests

Jun Rui Lee, Sadegh Emami, Michael David Hollins +5

Feature attribution (FA) methods are widely used in explainable AI (XAI) to help users understand how the inputs of a machine learning model contribute to its outputs. However, different FA models often provide disagreeing importance scores for the same model. In the absence of ground truth or in-depth knowledge about the inner workings of the model, it is often difficult to meaningfully determine which of the different FA methods produce more suitable explanations in different contexts. As a step towards addressing this issue, we introduce the open-source XAI-Units benchmark, specifically designed to evaluate FA methods against diverse types of model behaviours, such as feature interactions, cancellations, and discontinuous outputs. Our benchmark provides a set of paired datasets and models with known internal mechanisms, establishing clear expectations for desirable attribution scores. Accompanied by a suite of built-in evaluation metrics, XAI-Units streamlines systematic experimentation and reveals how FA methods perform against distinct, atomic kinds of model reasoning, similar to unit tests in software engineering. Crucially, by using procedurally generated models tied to synthetic datasets, we pave the way towards an objective and reliable comparison of FA methods.

Counterfactual explanations play a pivotal role in explainable artificial intelligence (XAI) by offering intuitive, human-understandable alternatives that elucidate machine learning model decisions. Despite their significance, existing methods for generating counterfactuals often require constant access to the predictive model, involve computationally intensive optimization for each instance and lack the flexibility to adapt to new user-defined constraints without retraining. In this paper, we propose DiCoFlex, a novel model-agnostic, conditional generative framework that produces multiple diverse counterfactuals in a single forward pass. Leveraging conditional normalizing flows trained solely on labeled data, DiCoFlex addresses key limitations by enabling real-time user-driven customization of constraints such as sparsity and actionability at inference time. Extensive experiments on standard benchmark datasets show that DiCoFlex outperforms existing methods in terms of validity, diversity, proximity, and constraint adherence, making it a practical and scalable solution for counterfactual generation in sensitive decision-making domains.

Recent efforts to develop trustworthy AI systems have increased interest in learning problems with explicit requirements, or constraints. In deep learning, however, such problems are often handled through fixed weighted-sum penalization: the constraints are added to the task loss with fixed coefficients, and the resulting scalarized objective is minimized. This position paper argues that fixed penalization is often ill-suited for deep learning problems with non-negotiable requirements for several reasons. First, in non-convex settings, the penalized and constrained problems are generally not equivalent, so solving the former need not solve the latter. Second, fixed penalization weakens hard requirements into soft penalties to be traded off against task performance. Third, choosing penalty coefficients to indirectly solve the constrained problem often involves costly trial and error, because changing them alters the penalized objective itself, and hence can mean solving the wrong problem altogether. We therefore argue that, when a deep learning problem specifies non-negotiable requirements, the constrained formulation itself should be the starting point, not the surrogate problem defined by fixed penalization. The appropriate solution strategy should then be chosen based on the problem's structure and scale.

As artificial intelligence (AI) regulations evolve and the regulatory landscape develops and becomes more complex, ensuring compliance with ethical guidelines and legal frameworks remains a challenge for AI developers. This paper introduces an AI-driven self-assessment chatbot designed to assist users in navigating the European Union AI Act and related standards. Leveraging a Retrieval-Augmented Generation (RAG) framework, the chatbot enables real-time, context-aware compliance verification by retrieving relevant regulatory texts and providing tailored guidance. By integrating both public and proprietary standards, it streamlines regulatory adherence, reduces complexity, and fosters responsible AI development. The paper explores the chatbot's architecture, comparing naive and graph-based RAG models, and discusses its potential impact on AI governance.

Access Controls Will Solve the Dual-Use Dilemma

Evžen Wybitul

AI safety systems face the dual-use dilemma. It is unclear whether to answer dual-use requests, since the same query could be either harmless or harmful depending on who made it and why. To make better decisions, such systems would need to examine requests' real-world context, but currently, they lack access to this information. Instead, they sometimes end up making arbitrary choices that result in refusing legitimate queries and allowing harmful ones, which hurts both utility and safety. To address this, we propose a conceptual framework based on access controls where only verified users can access dual-use outputs. We describe the framework's components, analyse its feasibility, and explain how it addresses both over-refusals and under-refusals. While only a high-level proposal, our work takes the first step toward giving model providers more granular tools for managing dual-use content. Such tools would enable users to access more capabilities without sacrificing safety, and offer regulators new options for targeted policies.

Deep learning models have shown promise in lung pathology detection from chest X-rays, but widespread clinical adoption remains limited due to opaque model decision-making. In prior work, we introduced ClinicXAI, a human-centric, expert-guided concept bottleneck model (CBM) designed for interpretable lung cancer diagnosis. We now extend that approach and present XpertXAI, a generalizable expert-driven model that preserves human-interpretable clinical concepts while scaling to detect multiple lung pathologies. Using a high-performing InceptionV3-based classifier and a public dataset of chest X-rays with radiology reports, we compare XpertXAI against leading post-hoc explainability methods and an unsupervised CBM, XCBs. We assess explanations through comparison with expert radiologist annotations and medical ground truth. Although XpertXAI is trained for multiple pathologies, our expert validation focuses on lung cancer. We find that existing techniques frequently fail to produce clinically meaningful explanations, omitting key diagnostic features and disagreeing with radiologist judgments. XpertXAI not only outperforms these baselines in predictive accuracy but also delivers concept-level explanations that better align with expert reasoning. While our focus remains on explainability in lung cancer detection, this work illustrates how human-centric model design can be effectively extended to broader diagnostic contexts - offering a scalable path toward clinically meaningful explainable AI in medical diagnostics.

Cooperation between the United States and China, the world's leading artificial intelligence (AI) powers, is crucial for effective global AI governance and responsible AI development. Although geopolitical tensions have emphasized areas of conflict, in this work, we identify potential common ground for productive dialogue by conducting a systematic analysis of more than 40 primary AI policy and corporate governance documents from both nations. Specifically, using an adapted version of the AI Governance and Regulatory Archive (AGORA) - a comprehensive repository of global AI governance documents - we analyze these materials in their original languages to identify areas of convergence in (1) sociotechnical risk perception and (2) governance approaches. We find strong and moderate overlap in several areas such as on concerns about algorithmic transparency, system reliability, agreement on the importance of inclusive multi-stakeholder engagement, and AI's role in enhancing safety. These findings suggest that despite strategic competition, there exist concrete opportunities for bilateral U.S.-China cooperation in the development of responsible AI. Thus, we present recommendations for furthering diplomatic dialogues that can facilitate such cooperation. Our analysis contributes to understanding how different international governance frameworks might be harmonized to promote global responsible AI development.

The need to explain predictive models is well-established in modern machine learning. However, beyond model interpretability, understanding pre-processing methods is equally essential. Understanding how data modifications impact model performance improvements and potential biases and promoting a reliable pipeline is mandatory for developing robust machine learning solutions. Isolation Forest (iForest) is a widely used technique for outlier detection that performs well. Its effectiveness increases with the number of tree-based learners. However, this also complicates the explanation of outlier selection and the decision boundaries for inliers. This research introduces a novel Explainable AI (XAI) method, tackling the problem of global explainability. In detail, it aims to offer a global explanation for outlier detection to address its opaque nature. Our approach is based on the Decision Predicate Graph (DPG), which clarifies the logic of ensemble methods and provides both insights and a graph-based metric to explain how samples are identified as outliers using the proposed Inlier-Outlier Propagation Score (IOP-Score). Our proposal enhances iForest's explainability and provides a comprehensive view of the decision-making process, detailing which features contribute to outlier identification and how the model utilizes them. This method advances the state-of-the-art by providing insights into decision boundaries and a comprehensive view of holistic feature usage in outlier identification. -- thus promoting a fully explainable machine learning pipeline.