Research Papers

As machine learning models and autonomous agents are increasingly deployed in high-stakes, real-world domains such as healthcare, security, finance, and robotics, the need for transparent and trustworthy explanations has become critical. To ensure end-to-end transparency of AI decisions, we need models that are not only accurate but also fully explainable and human-tunable. We introduce BACON, a novel framework for automatically training explainable AI models for decision making problems using graded logic. BACON achieves high predictive accuracy while offering full structural transparency and precise, logic-based symbolic explanations, enabling effective human-AI collaboration and expert-guided refinement. We evaluate BACON with a diverse set of scenarios: classic Boolean approximation, Iris flower classification, house purchasing decisions and breast cancer diagnosis. In each case, BACON provides high-performance models while producing compact, human-verifiable decision logic. These results demonstrate BACON's potential as a practical and principled approach for delivering crisp, trustworthy explainable AI.

Ensuring the reliability and safety of machine learning models in open-world deployment is a central challenge in AI safety. This thesis develops both algorithmic and theoretical foundations to address key reliability issues arising from distributional uncertainty and unknown classes, from standard neural networks to modern foundation models like large language models (LLMs). Traditional learning paradigms, such as empirical risk minimization (ERM), assume no distribution shift between training and inference, often leading to overconfident predictions on out-of-distribution (OOD) inputs. This thesis introduces novel frameworks that jointly optimize for in-distribution accuracy and reliability to unseen data. A core contribution is the development of an unknown-aware learning framework that enables models to recognize and handle novel inputs without labeled OOD data. We propose new outlier synthesis methods, VOS, NPOS, and DREAM-OOD, to generate informative unknowns during training. Building on this, we present SAL, a theoretical and algorithmic framework that leverages unlabeled in-the-wild data to enhance OOD detection under realistic deployment conditions. These methods demonstrate that abundant unlabeled data can be harnessed to recognize and adapt to unforeseen inputs, providing formal reliability guarantees. The thesis also extends reliable learning to foundation models. We develop HaloScope for hallucination detection in LLMs, MLLMGuard for defending against malicious prompts in multimodal models, and data cleaning methods to denoise human feedback used for better alignment. These tools target failure modes that threaten the safety of large-scale models in deployment. Overall, these contributions promote unknown-aware learning as a new paradigm, and we hope it can advance the reliability of AI systems with minimal human efforts.

Large language models (LLMs) offer unprecedented and growing capabilities, but also introduce complex safety and security challenges that resist conventional risk management. While conventional probabilistic risk analysis (PRA) requires exhaustive risk enumeration and quantification, the novelty and complexity of these systems make PRA impractical, particularly against adaptive adversaries. Previous research found that risk management in various fields of engineering such as nuclear or civil engineering is often solved by generic (i.e. field-agnostic) strategies such as event tree analysis or robust designs. Here we show how emerging risks in LLM-powered systems could be met with 100+ of these non-probabilistic strategies to risk management, including risks from adaptive adversaries. The strategies are divided into five categories and are mapped to LLM security (and AI safety more broadly). We also present an LLM-powered workflow for applying these strategies and other workflows suitable for solution architects. Overall, these strategies could contribute (despite some limitations) to security, safety and other dimensions of responsible AI.

Graph neural networks have become the de facto model for learning from structured data. However, the decision-making process of GNNs remains opaque to the end user, which undermines their use in safety-critical applications. Several explainable AI techniques for graphs have been developed to address this major issue. Focusing on graph classification, these explainers identify subgraph motifs that explain predictions. Therefore, a robust benchmarking of graph explainers is required to ensure that the produced explanations are of high quality, i.e., aligned with the GNN's decision process. However, current graph-XAI benchmarks are limited to simplistic synthetic datasets or a few real-world tasks curated by domain experts, hindering rigorous and reproducible evaluation, and consequently stalling progress in the field. To overcome these limitations, we propose a method to automate the construction of graph XAI benchmarks from generic graph classification datasets. Our approach leverages the Weisfeiler-Leman color refinement algorithm to efficiently perform approximate subgraph matching and mine class-discriminating motifs, which serve as proxy ground-truth class explanations. At the same time, we ensure that these motifs can be learned by GNNs because their discriminating power aligns with WL expressiveness. This work also introduces the OpenGraphXAI benchmark suite, which consists of 15 ready-made graph-XAI datasets derived by applying our method to real-world molecular classification datasets. The suite is available to the public along with a codebase to generate over 2,000 additional graph-XAI benchmarks. Finally, we present a use case that illustrates how the suite can be used to assess the effectiveness of a selection of popular graph explainers, demonstrating the critical role of a sufficiently large benchmark collection for improving the significance of experimental results.

The issues of AI risk and AI safety are becoming critical as the prospect of artificial general intelligence (AGI) looms larger. The emergence of extremely large and capable generative models has led to alarming predictions and created a stir from boardrooms to legislatures. As a result, AI alignment has emerged as one of the most important areas in AI research. The goal of this position paper is to argue that the currently dominant vision of AGI in the AI and machine learning (AI/ML) community needs to evolve, and that expectations and metrics for its safety must be informed much more by our understanding of the only existing instance of general intelligence, i.e., the intelligence found in animals, and especially in humans. This change in perspective will lead to a more realistic view of the technology, and allow for better policy decisions.

Federated Learning (FL) has emerged as a powerful paradigm for collaborative model training while keeping client data decentralized and private. However, it is vulnerable to Data Reconstruction Attacks (DRA) such as "LoKI" and "Robbing the Fed", where malicious models sent from the server to the client can reconstruct sensitive user data. To counter this, we introduce DRArmor, a novel defense mechanism that integrates Explainable AI with targeted detection and mitigation strategies for DRA. Unlike existing defenses that focus on the entire model, DRArmor identifies and addresses the root cause (i.e., malicious layers within the model that send gradients with malicious intent) by analyzing their contribution to the output and detecting inconsistencies in gradient values. Once these malicious layers are identified, DRArmor applies defense techniques such as noise injection, pixelation, and pruning to these layers rather than the whole model, minimizing the attack surface and preserving client data privacy. We evaluate DRArmor's performance against the advanced LoKI attack across diverse datasets, including MNIST, CIFAR-10, CIFAR-100, and ImageNet, in a 200-client FL setup. Our results demonstrate DRArmor's effectiveness in mitigating data leakage, achieving high True Positive and True Negative Rates of 0.910 and 0.890, respectively. Additionally, DRArmor maintains an average accuracy of 87%, effectively protecting client privacy without compromising model performance. Compared to existing defense mechanisms, DRArmor reduces the data leakage rate by 62.5% with datasets containing 500 samples per client.

Algorithmic bias has been the subject of much recent controversy. To clarify what is at stake and to make progress resolving the controversy, a better understanding of the concepts involved would be helpful. The discussion here focuses on the disputed claim that algorithms themselves cannot be biased. To clarify this claim we need to know what kind of thing 'algorithms themselves' are, and to disambiguate the several meanings of 'bias' at play. This further involves showing how bias of moral import can result from statistical biases, and drawing connections to previous conceptual work about political artifacts and oppressive things. Data bias has been identified in domains like hiring, policing and medicine. Examples where algorithms themselves have been pinpointed as the locus of bias include recommender systems that influence media consumption, academic search engines that influence citation patterns, and the 2020 UK algorithmically-moderated A-level grades. Recognition that algorithms are a kind of thing that can be biased is key to making decisions about responsibility for harm, and preventing algorithmically mediated discrimination.

This article introduces WebXAII, an open-source web framework designed to facilitate research on human interaction with eXplainable Artificial Intelligence (XAI) systems. The field of XAI is rapidly expanding, driven by the growing societal implications of the widespread adoption of AI (and in particular machine learning) across diverse applications. Researchers who study the interaction between humans and XAI techniques typically develop ad hoc interfaces in order to conduct their studies. These interfaces are usually not shared alongside the results of the studies, which limits their reusability and the reproducibility of experiments. In response, we design and implement WebXAII, a web-based platform that can embody full experimental protocols, meaning that it can present all aspects of the experiment to human participants and record their responses. The experimental protocols are translated into a composite architecture of generic views and modules, which offers a lot of flexibility. The architecture is defined in a structured configuration file, so that protocols can be implemented with minimal programming skills. We demonstrate that WebXAII can effectively embody relevant protocols, by reproducing the protocol of a state-of-the-art study of the literature.

As the field of healthcare increasingly adopts artificial intelligence, it becomes important to understand which types of explanations increase transparency and empower users to develop confidence and trust in the predictions made by machine learning (ML) systems. In shared decision-making scenarios where doctors cooperate with ML systems to reach an appropriate decision, establishing mutual trust is crucial. In this paper, we explore different approaches to generating explanations in eXplainable AI (XAI) and make their underlying arguments explicit so that they can be evaluated by medical experts. In particular, we present the findings of a user study conducted with physicians to investigate their perceptions of various types of AI-generated explanations in the context of diagnostic decision support. The study aims to identify the most effective and useful explanations that enhance the diagnostic process. In the study, medical doctors filled out a survey to assess different types of explanations. Further, an interview was carried out post-survey to gain qualitative insights on the requirements of explanations incorporated in diagnostic decision support. Overall, the insights gained from this study contribute to understanding the types of explanations that are most effective.

ARTO is a project combining collaborative robots (cobots) and Artificial Intelligence (AI) to automate functional test procedures for civilian and military aircraft certification. This paper proposes a Deep Learning (DL) and eXplainable AI (XAI) approach, equipping ARTO with interaction analysis capabilities to verify and validate the operations on cockpit components. During these interactions, forces, torques, and end effector poses are recorded and preprocessed to filter disturbances caused by low performance force controllers and embedded Force Torque Sensors (FTS). Convolutional Neural Networks (CNNs) then classify the cobot actions as Success or Fail, while also identifying and reporting the causes of failure. To improve interpretability, Grad CAM, an XAI technique for visual explanations, is integrated to provide insights into the models decision making process. This approach enhances the reliability and trustworthiness of the automated testing system, facilitating the diagnosis and rectification of errors that may arise during testing.

Explainable AI (XAI) research has traditionally focused on rational users, aiming to improve understanding and reduce cognitive biases. However, emotional factors play a critical role in how explanations are perceived and processed. Prior work shows that prior and task-generated emotions can negatively impact the understanding of explanation. Building on these insights, we propose a three-stage model for emotion-sensitive explanation grounding: (1) emotional or epistemic arousal, (2) understanding, and (3) agreement. This model provides a conceptual basis for developing XAI systems that dynamically adapt explanation strategies to users emotional states, ultimately supporting more effective and user-centered decision-making.

As society becomes increasingly reliant on artificial intelligence, the need to mitigate risk and harm is paramount. In response, researchers and practitioners have developed tools to detect and reduce undesired bias, commonly referred to as fairness tools. Many of these tools are publicly available for free use and adaptation. While the growing availability of such tools is promising, little is known about the broader landscape beyond well-known examples like AI Fairness 360 and Fairlearn. Because fairness is an ongoing concern, these tools must be built for long-term sustainability. Using an existing set of fairness tools as a reference, we systematically searched GitHub and identified 50 related projects. We then analyzed various aspects of their repositories to assess community engagement and the extent of ongoing maintenance. Our findings show diverse forms of engagement with these tools, suggesting strong support for open-source development. However, we also found significant variation in how well these tools are maintained. Notably, 53 percent of fairness projects become inactive within the first three years. By examining sustainability in fairness tooling, we aim to promote more stability and growth in this critical area.