Research Papers

A central challenge in the development of nanophotonic structures is identifying the optimal design for a target functionality, and understanding the physical mechanisms that enable the optimized device's capabilities. Previously investigated design methods for nanophotonic structures, including both conventional optimization approaches as well as nascent machine learning (ML) strategies, have made progress, yet they remain 'black boxes' that lack explanations for their predictions. Here we demonstrate that convolutional neural networks (CNN) trained to predict the electromagnetic response of classes of metal-dielectric-metal metamaterials, including complex freeform designs, can be explained to reveal deeper insights into the underlying physics of nanophotonic structures. Using an explainable AI (XAI) approach, we show that we can identify the importance of specific spatial regions of a nanophotonic structure for the presence or lack of an absorption peak. Our results highlight that ML strategies can be used for physics discovery, as well as design optimization, in optics and photonics.

A Set of Distinct Facial Traits Learned by Machines Is Not Predictive of Appearance Bias in the Wild

Ryan Steed, Aylin Caliskan

Research in social psychology has shown that people's biased, subjective judgments about another's personality based solely on their appearance are not predictive of their actual personality traits. But researchers and companies often utilize computer vision models to predict similarly subjective personality attributes such as "employability." We seek to determine whether state-of-the-art, black box face processing technology can learn human-like appearance biases. With features extracted with FaceNet, a widely used face recognition framework, we train a transfer learning model on human subjects' first impressions of personality traits in other faces as measured by social psychologists. We find that features extracted with FaceNet can be used to predict human appearance bias scores for deliberately manipulated faces but not for randomly generated faces scored by humans. Additionally, in contrast to work with human biases in social psychology, the model does not find a significant signal correlating politicians' vote shares with perceived competence bias. With Local Interpretable Model-Agnostic Explanations (LIME), we provide several explanations for this discrepancy. Our results suggest that some signals of appearance bias documented in social psychology are not embedded by the machine learning techniques we investigate. We shed light on the ways in which appearance bias could be embedded in face processing technology and cast further doubt on the practice of predicting subjective traits based on appearances.

TX-Ray: Quantifying and Explaining Model-Knowledge Transfer in (Un-)Supervised NLP

Nils Rethmeier, Vageesh Kumar Saxena, Isabelle Augenstein

While state-of-the-art NLP explainability (XAI) methods focus on explaining per-sample decisions in supervised end or probing tasks, this is insufficient to explain and quantify model knowledge transfer during (un-)supervised training. Thus, for TX-Ray, we modify the established computer vision explainability principle of 'visualizing preferred inputs of neurons' to make it usable transfer analysis and NLP. This allows one to analyze, track and quantify how self- or supervised NLP models first build knowledge abstractions in pretraining (1), and then transfer these abstractions to a new domain (2), or adapt them during supervised fine-tuning (3). TX-Ray expresses neurons as feature preference distributions to quantify fine-grained knowledge transfer or adaptation and guide human analysis. We find that, similar to Lottery Ticket based pruning, TX-Ray based pruning can improve test set generalization and that it can reveal how early stages of self-supervision automatically learn linguistic abstractions like parts-of-speech.

The social implications of algorithmic decision-making in sensitive contexts have generated lively debates among multiple stakeholders, such as moral and political philosophers, computer scientists, and the public. Yet, the lack of a common language and a conceptual framework for an appropriate bridging of the moral, technical, and political aspects of the debate prevents the discussion to be as effective as it can be. Social scientists and psychologists are contributing to this debate by gathering a wealth of empirical data, yet a philosophical analysis of the social implications of algorithmic decision-making remains comparatively impoverished. In attempting to address this lacuna, this paper argues that a hierarchy of different types of explanations for why and how an algorithmic decision outcome is achieved can establish the relevant connection between the moral and technical aspects of algorithmic decision-making. In particular, I offer a multi-faceted conceptual framework for the explanations and the interpretations of algorithmic decisions, and I claim that this framework can lay the groundwork for a focused discussion among multiple stakeholders about the social implications of algorithmic decision-making, as well as AI governance and ethics more generally.

This work showcases a new approach for causal discovery by leveraging user experiments and recent advances in photo-realistic image editing, demonstrating a potential of identifying causal factors and understanding complex systems counterfactually. We introduce the beauty learning problem as an example, which has been discussed metaphysically for centuries and been proved exists, is quantifiable, and can be learned by deep models in our recent paper, where we utilize a natural image generator coupled with user studies to infer causal effects from facial semantics to beauty outcomes, the results of which also align with existing empirical studies. We expect the proposed framework for a broader application in causal inference.

Predictive geometric models deliver excellent results for many Machine Learning use cases. Despite their undoubted performance, neural predictive algorithms can show unexpected degrees of instability and variance, particularly when applied to large datasets. We present an approach to measure changes in geometric models with respect to both output consistency and topological stability. Considering the example of a recommender system using word2vec, we analyze the influence of single data points, approximation methods and parameter settings. Our findings can help to stabilize models where needed and to detect differences in informational value of data points on a large scale.

Math for AI safety: an invitation for mathematicians

Lionel Levine

Artificial intelligence threatens to outrun human understanding and control. New mathematics is needed to design AI that is legible, steerable, and cooperative with humanity. I organize this invitation by mathematical field, so you can turn straight to your own: logic and game theory for cooperation; probability for agency and world-models; algebra and representation theory for learned features; analysis and geometry for generalization and training dynamics. Each section ends with an open problem that is accessible to a working mathematician with no prior experience in AI safety.

The rapid growth of Distributed Energy Resources (DERs) has significantly expanded the cyber attack surface of modern power grids. Furthermore, increasing sophistication in attack techniques demands anomaly detection systems (ADS) that are accurate, interpretable, and reliable to support DER cybersecurity. While ML-based ADS provide strong detection capabilities, their black-box nature reduces operator trust and limits Security Operation Center's (SOC) ability to effectively interpret alerts and respond, highlighting the need for explainable Artificial Intelligence (XAI) to ensure transparency and operational confidence. This paper presents an XAI-based anomaly detection framework tailored for DER networks (ExCYDER). The proposed framework uses a self-verifying mechanism that validates ADS alerts to ensure trustworthy decision-making. ExCYDER combines LightGBM with SHAP to check whether each model decision aligns with its feature-attribution evidence, allowing the system to confirm that its internal reasoning is consistent and reliable. Experiments on a realistic DNP3 dataset achieved over 98% detection accuracy, an average rule--SHAP consistency of 44.6%, a SHAP latency of 14.5 ms per alert, and a confidence deviation within 5%, demonstrating stable verification behavior with minimal computational overhead. The framework distinguished between coherent and inconsistent alerts without compromising detection accuracy, demonstrating that integrated verification within XAI-based ADS enhances interpretability, auditability, and operational robustness for DER-focused SOCs.

Interpretability and fairness are two of the most emphasized dimensions in trustworthy artificial intelligence (AI). Various explainable AI methods have been introduced to improve interpretability. This paper focuses on neural network (NN)-based generalized additive models (GAMs), a class of self-interpretable models. While most existing research has prioritized improving the accuracy of NN-based GAMs, their interpretability remains largely underexplored. To address this gap, this paper introduces explicit quantitative metrics for evaluating the interpretability of NN-based GAMs, empirically examines their effectiveness, and explores strategies for improving interpretability within these models. In addition, the simultaneous and explicit optimization of both interpretability and fairness, along with their trade-offs and the underlying reasons, remains underexplored. To address this, we propose a multi-objective neural basis model (MONBM) framework based on multi-objective evolutionary learning to consider accuracy, interpretability, and fairness simultaneously. A partial retraining strategy is further developed to facilitate the practical application of evolutionary multi-objective optimization to deep model architectures. Based on MONBM, this paper reveals the complex relationships between these dimensions and the reasons behind these intricate relationships. This analysis demonstrates how multi-objective optimization can be combined with self-interpretable models to reveal relationships among trustworthiness objectives. In addition, MONBM obtains a set of models with different trade-offs between dimensions, and the competitiveness of the approach is validated by comparing it with state-of-the-art methods.

AI oversight methods rely on ground truth for validation, but what constitutes appropriate AI behavior is contested. This leaves evaluation of moral reasoning in LLMs and debate-based oversight implicitly avoiding realistic ambiguity. We investigate an alternative standard designed to function despite such ambiguity: structural quality of the defence a model can mount for its verdicts in response to critical questions, measured through a four-phase dialectical protocol grounded in Walton's theory of argumentation schemes and Govier's criteria for argument cogency. The protocol is adaptive to different frames of reasoning, extends beyond multiple-choice framing, and treats both the reasoning that precedes a verdict and its post-hoc justification. Across nine frontier models and 200 high-ambiguity MoralChoice items -- $6,778$ judge-scored cells, validated against $89.6\%$ inter-judge agreement on the binary failure judgment -- models defend their reasoning well above the rubric minimum on every dimension. Failure mass concentrates on grounds and sufficiency, and correlates with epistemic hedging rather than argument length. Reasoning is better defended than post-hoc justification, on every model and every Govier dimension. The scheme a model presents in its justification differs from the one it reasoned with on a substantial share of dilemmas ($\geq 20\%$ per model), despite value-based practical reasoning dominating both tracks. The protocol catches strictly indefensible defences (self-contradiction, false premises), and it surfaces difficulties in characterizing the role of retraction in AI alignment, suggesting a need for more situated evaluations.

Interpretability is critical for machine learning models deployed in scientific space missions such as ESA's Ariel, where ground truth is unavailable during operations and physical plausibility must be assessed. While most explainable AI methods focus on feature attribution, this work investigates training data attribution through influence functions and introduces three key contributions for operational spectroscopy pipelines. First, influence is reformulated in terms of prediction rather than loss, enabling label-free deployment. Second, by leveraging the closed-form ridge solution of an Extreme Learning Machine, infinitesimal prediction influence is efficiently computed. Third, an influence-based conservative error proxy is derived by propagating training residuals through the influence sensitivities. Evaluated against simulated spectra, the proposed proxy correlates strongly with scale and shape-based spectral errors. Furthermore, influence functions enable the identification of the most influential samples and the approximation of the most harmful ones. Together, these results suggest that this approach can serve as an operational framework for scientific machine learning.

SHAP and LIME are now standard tools for interpreting black-box predictions, yet their outputs can vary substantially when the input is perturbed by small amounts of noise--a problem we observed firsthand in our previous work on food security in Madagascar (Ralinirina et al., 2025). This variability raises the question of whether such explanations can be trusted at all. We address it by constructing an auditing protocol that measures two properties of any post-hoc explainer: robustness (how stable the explanation is under input perturbation) and fidelity (whether the features deemed important actually drive the model's prediction). These two quantities are combined into a single Trust Score. We run the protocol on a multi-sectoral dataset from Madagascar (83 features, 253 records, 4 malnutrition classes) using three classifiers and two explainers, plus their regularized counterparts. The results are sobering: models with AUC above 0.99 can produce numerically degenerate or flatly uninformative explanations, and fidelity scores lose discriminative power when the model is overfitted. These findings suggest that auditing XAI outputs is not optional but necessary, particularly when they inform decisions in sensitive domains.