Abstract
Gender bias in vision-language foundation models (VLMs) raises concerns about their safe deployment and is typically evaluated using benchmarks with gender annotations on real-world images. However, as these benchmarks often contain spurious correlations between gender and non-gender features, such as objects and backgrounds, we identify a critical oversight in gender bias evaluation: Do spurious features distort gender bias evaluation? To address this question, we systematically perturb non-gender features across four widely used benchmarks (COCO-gender, FACET, MIAP, and PHASE) and various VLMs to quantify their impact on bias evaluation. Our findings reveal that even minimal perturbations, such as masking just 10% of objects or weakly blurring backgrounds, can dramatically alter bias scores, shifting metrics by up to 175% in generative VLMs and 43% in CLIP variants. This suggests that current bias evaluations often reflect model responses to spurious features rather than gender bias, undermining their reliability. Since creating spurious feature-free benchmarks is fundamentally challenging, we recommend reporting bias metrics alongside feature-sensitivity measurements to enable a more reliable bias assessment.

AI Insights

• Detector choice can swing bias scores, so benchmarking must account for the specific object detector used.
• Stronger spurious cues (e.g., prominent objects) amplify bias shifts, a trend quantified by the Acc bin correlation.
• Ensemble or self‑supervised detection pipelines promise to disentangle detector noise from genuine bias signals.
• Reporting a feature‑sensitivity curve alongside bias metrics offers a clearer picture of model robustness.
• Spurious Feature: an irrelevant visual cue that nonetheless sways model predictions.
• Bias Measurement: a quantitative score (accuracy, precision, etc.) that captures systematic disparities.
• Key reading: “Measuring Bias in Visual Language Models” and “Self‑Supervised Learning for Reducing Bias in AI Systems” for deeper dives.

Abstract
The simplicity Bias (SB) of neural networks, i.e.\ their tendency to represent simple functions, is a key factor in their generalization capabilities. Recent studies show that an excessive SB may harm performance on complex tasks, and the need for this bias varies across tasks. Many of these studies focus on simple models or synthetic tasks. It remains challenging to measure the SB in large models and little is known about the relevance of the SB to various image classification tasks. In this paper, we investigate the relationship between the SB in CLIP models and their performance across image classification tasks. First, we theoretically analyze the potential limitation of existing measures of complexity that have been used to characterize small models. To address this, we propose a frequency-aware measure capturing finer-grained SB differences. We validate this measure on CLIP models subjected to two recent SB-modulation methods, demonstrating that it is more informative and consistent than previous measures. Second, we examine the relation between the SB of those models and their performance across a range of image classification tasks, including zero-shot and fine-tuning settings. These experiments reveal a range of behaviors. For example, a stronger SB correlates with a better performance on OOD generalization than on adversarial robustness. These results highlight the benefits of aligning a model's inductive biases with the characteristics of the target task.

Abstract
Data exfiltration is a growing problem for business who face costs related to the loss of confidential data as well as potential extortion. This work presents a simple game theoretic model of network data exfiltration. In the model, the attacker chooses the exfiltration route and speed, and the defender selects monitoring thresholds to detect unusual activity. The attacker is rewarded for exfiltrating data, and the defender tries to minimize the costs of data loss and of responding to alerts.

AI Insights

• Link‑level distributions let defenders quantify confidence via standard deviation, turning traffic variance into alert metrics.
• Adding a detection‑accuracy probability when an alert fires can shift the attacker–defender equilibrium.
• Structural equilibrium methods scale better than exhaustive search as the network grows to many devices.
• The game adapts to diverse exfiltration channels, from covert DNS tunneling to encrypted VPN leaks.
• Extending to multi‑device topologies quickly inflates the state space, underscoring the need for abstraction.
• The POMDP study “Data Exfiltration Detection and Prevention” offers a complementary probabilistic framework.
• “Decision and Game Theory for Security” supplies a solid theoretical base for more sophisticated exfiltration games.

Abstract
The emergence of synthetic data for privacy protection, training data generation, or simply convenient access to quasi-realistic data in any shape or volume complicates the concept of ground truth. Synthetic data mimic real-world observations, but do not refer to external features. This lack of a representational relationship, however, not prevent researchers from using synthetic data as training data for AI models and ground truth repositories. It is claimed that the lack of data realism is not merely an acceptable tradeoff, but often leads to better model performance than realistic data: compensate for known biases, prevent overfitting and support generalization, and make the models more robust in dealing with unexpected outliers. Indeed, injecting noisy and outright implausible data into training sets can be beneficial for the model. This greatly complicates usual assumptions based on which representational accuracy determines data fidelity (garbage in - garbage out). Furthermore, ground truth becomes a self-referential affair, in which the labels used as a ground truth repository are themselves synthetic products of a generative model and as such not connected to real-world observations. My paper examines how ML researchers and practitioners bootstrap ground truth under such paradoxical circumstances without relying on the stable ground of representation and real-world reference. It will also reflect on the broader implications of a shift from a representational to what could be described as a mimetic or iconic concept of data.

Abstract
Fairness-aware classification requires balancing performance and fairness, often intensified by intersectional biases. Conflicting fairness definitions further complicate the task, making it difficult to identify universally fair solutions. Despite growing regulatory and societal demands for equitable AI, popular toolkits offer limited support for exploring multi-dimensional fairness and related trade-offs. To address this, we present mmm-fair, an open-source toolkit leveraging boosting-based ensemble approaches that dynamically optimizes model weights to jointly minimize classification errors and diverse fairness violations, enabling flexible multi-objective optimization. The system empowers users to deploy models that align with their context-specific needs while reliably uncovering intersectional biases often missed by state-of-the-art methods. In a nutshell, mmm-fair uniquely combines in-depth multi-attribute fairness, multi-objective optimization, a no-code, chat-based interface, LLM-powered explanations, interactive Pareto exploration for model selection, custom fairness constraint definition, and deployment-ready models in a single open-source toolkit, a combination rarely found in existing fairness tools. Demo walkthrough available at: https://youtu.be/_rcpjlXFqkw.

AI Insights

• mmm‑fair only handles tabular data and requires users to manually list protected attributes, limiting raw image or text use.
• Its explanation engine relies on external LLMs, so an API key is mandatory and usage costs may rise.
• Designed for small‑to‑medium studies, it lacks causal inference, longitudinal modeling, and large‑scale deployment.
• Planned extensions: desktop/cloud interface, local‑model support, hybrid explanations, and vision/text modality expansion.
• Key reading: Barocas et al.’s “Fairness and Machine Learning” and Friedler et al.’s (Im)possibility of Fairness paper highlight why no single metric satisfies all stakeholders.
• Definitions: Fairness‑aware ML mitigates bias; multi‑fairness trade‑offs balance accuracy against multiple fairness metrics.

Abstract
Evaluating network-wide fairness is challenging because it is not a static property but one highly sensitive to Service Level Agreement (SLA) parameters. This paper introduces a complete analytical framework to transform fairness evaluation from a single-point measurement into a proactive engineering discipline centered on a predictable sensitivity landscape. Our framework is built upon a QoE-Imbalance metric whose form is not an ad-hoc choice, but is uniquely determined by a set of fundamental axioms of fairness, ensuring its theoretical soundness. To navigate the fairness landscape across the full spectrum of service demands, we first derive a closed-form covariance rule. This rule provides an interpretable, local compass, expressing the fairness gradient as the covariance between a path's information-theoretic importance and its parameter sensitivity. We then construct phase diagrams to map the global landscape, revealing critical topological features such as robust "stable belts" and high-risk "dangerous wedges". Finally, an analysis of the landscape's curvature yields actionable, topology-aware design rules, including an optimal "Threshold-First" tuning strategy. Ultimately, our framework provides the tools to map, interpret, and navigate the landscape of system sensitivity, enabling the design of more robust and resilient networks.

Abstract
From altruism to antagonism, fairness plays a central role in social interactions. But can we truly understand how fair someone is, especially without explicit knowledge of their preferences? We cast this challenge as a multi-agent inverse reinforcement learning problem, explicitly structuring rewards to reflect how agents value the welfare of others. We introduce novel Bayesian strategies, reasoning about the optimality of demonstrations and characterisation of equilibria in general-sum Markov games. Our experiments, spanning randomised environments and a collaborative cooking task, reveal that coherent notions of fairness can be reliably inferred from demonstrations. Furthermore, when isolating fairness components, we obtain a disentangled understanding of agents preferences. Crucially, we unveil that by placing agents in different groups, we can force them to exhibit new facets of their reward structures, cutting through ambiguity to answer the central question: who is being fair?

Abstract
This paper introduces PyFair, a formal framework for evaluating and verifying individual fairness of Deep Neural Networks (DNNs). By adapting the concolic testing tool PyCT, we generate fairness-specific path constraints to systematically explore DNN behaviors. Our key innovation is a dual network architecture that enables comprehensive fairness assessments and provides completeness guarantees for certain network types. We evaluate PyFair on 25 benchmark models, including those enhanced by existing bias mitigation techniques. Results demonstrate PyFair's efficacy in detecting discriminatory instances and verifying fairness, while also revealing scalability challenges for complex models. This work advances algorithmic fairness in critical domains by offering a rigorous, systematic method for fairness testing and verification of pre-trained DNNs.

AI Insights

• Fairify PyCT builds a 2‑DNN by duplicating each layer’s weights, enabling symmetry‑aware fairness checks.
• In benchmarks, it finds bias in 18 of 25 models, outperforming PyFair by ~30 %.
• Its concolic engine generates path constraints that isolate protected‑attribute violations, revealing subtle discrimination.
• Runtime scales quadratically with depth, making it resource‑hungry for very deep networks.
• Results differ by protected attribute; gender and age yield higher false‑positive rates than race.
• For deeper dives, see “Fairness in Machine Learning: A Survey” and the repo at https://github.com/fairlearn/fairlearn.
• The tool can be integrated into CI pipelines for continuous fairness monitoring during deployment.

Abstract
Ethical questions are discussed regularly in computer security. Still, researchers in computer security lack clear guidance on how to make, document, and assess ethical decisions in research when what is morally right or acceptable is not clear-cut. In this work, we give an overview of the discussion of ethical implications in current published work in computer security by reviewing all 1154 top-tier security papers published in 2024, finding inconsistent levels of ethics reporting with a strong focus of reporting institutional or ethics board approval, human subjects protection, and responsible disclosure, and a lack of discussion of balancing harms and benefits. We further report on the results of a semi-structured interview study with 24 computer security and privacy researchers (among whom were also: reviewers, ethics committee members, and/or program chairs) and their ethical decision-making both as authors and during peer review, finding a strong desire for ethical research, but a lack of consistency in considered values, ethical frameworks (if articulated), decision-making, and outcomes. We present an overview of the current state of the discussion of ethics and current de-facto standards in computer security research, and contribute suggestions to improve the state of ethics in computer security research.

AI Insights

• Only 3.9% of 1,154 top‑conference papers mention ethics, exposing a reporting gap.
• Ethics citations focus on 43% confidentiality, 34% human subjects, 29% IRB approval.
• Interviews with 24 researchers show reviewers rarely disclose ethics concerns in peer review.
• Only 19% of papers report IRB approval or exemption, revealing a compliance shortfall.
• Authors urge structured ethics training and resource hubs to empower responsible research.
• Study limits: self‑reported data and focus on top conferences may bias results.
• Curiously, the paper itself offers no concrete guidelines, underscoring the need for actionable frameworks.

Abstract
Contemporary robots are increasingly mimicking human social behaviours to facilitate interaction, such as smiling to signal approachability, or hesitating before taking an action to allow people time to react. Such techniques can activate a person's entrenched social instincts, triggering emotional responses as though they are interacting with a fellow human, and can prompt them to treat a robot as if it truly possesses the underlying life-like processes it outwardly presents, raising significant ethical questions. We engage these issues through the lens of informed consent: drawing upon prevailing legal principles and ethics, we examine how social robots can influence user behaviour in novel ways, and whether under those circumstances users can be appropriately informed to consent to these heightened interactions. We explore the complex circumstances of human-robot interaction and highlight how it differs from more familiar interaction contexts, and we apply legal principles relating to informed consent to social robots in order to reconceptualize the current ethical debates surrounding the field. From this investigation, we synthesize design goals for robot developers to achieve more ethical and informed human-robot interaction.

AI Insights

• Anthropomorphism inflates expectations, turning a simple robot into a perceived sentient companion, complicating consent.
• Deceptive cues—hesitation, smiling—activate human instincts, raising legal questions about manipulation.
• The paper maps informed consent onto HRI using tort and privacy doctrines, exposing regulatory gaps.
• Design guidelines: robots must disclose their artificial nature and limits before emotionally charged interactions.
• Eldercare and f-commerce are high‑stakes arenas where consent lapses can have severe consequences.
• Recommended reading: “Close Engagements with Artificial Companions” and a Harvard Law Review article on robotic agency.
• Informed consent in HRI is dynamic; it must evolve with robot sophistication and user familiarity.

Abstract
Sparse dictionary coding represents signals as linear combinations of a few dictionary atoms. It has been applied to images, time series, graph signals and multi-way spatio-temporal data by jointly employing temporal and spatial dictionaries. Data-agnostic analytical dictionaries, such as the discrete Fourier transform, wavelets and graph Fourier, have seen wide adoption due to efficient implementations and good practical performance. On the other hand, dictionaries learned from data offer sparser and more accurate solutions but require learning of both the dictionaries and the coding coefficients. This becomes especially challenging for multi-dictionary scenarios since encoding coefficients correspond to all atom combinations from the dictionaries. To address this challenge, we propose a low-rank coding model for 2-dictionary scenarios and study its data complexity. Namely, we establish a bound on the number of samples needed to learn dictionaries that generalize to unseen samples from the same distribution. We propose a convex relaxation solution, called AODL, whose exact solution we show also solves the original problem. We then solve this relaxation via alternating optimization between the sparse coding matrices and the learned dictionaries, which we prove to be convergent. We demonstrate its quality for data reconstruction and missing value imputation in both synthetic and real-world datasets. For a fixed reconstruction quality, AODL learns up to 90\% sparser solutions compared to non-low-rank and analytical (fixed) dictionary baselines. In addition, the learned dictionaries reveal interpretable insights into patterns present within the samples used for training.

Abstract
A ubiquitous feature of data of our era is their extra-large sizes and dimensions. Analyzing such high-dimensional data poses significant challenges, since the feature dimension is often much larger than the sample size. This thesis introduces robust and computationally efficient methods to address several common challenges associated with high-dimensional data. In my first manuscript, I propose a coherent approach to variable screening that accommodates nonlinear associations. I develop a novel variable screening method that transcends traditional linear assumptions by leveraging mutual information, with an intended application in neuroimaging data. This approach allows for accurate identification of important variables by capturing nonlinear as well as linear relationships between the outcome and covariates. Building on this foundation, I develop new optimization methods for sparse estimation using nonconvex penalties in my second manuscript. These methods address notable challenges in current statistical computing practices, facilitating computationally efficient and robust analyses of complex datasets. The proposed method can be applied to a general class of optimization problems. In my third manuscript, I contribute to robust modeling of high-dimensional correlated observations by developing a mixed-effects model based on Tsallis power-law entropy maximization and discussed the theoretical properties of such distribution. This model surpasses the constraints of conventional Gaussian models by accommodating a broader class of distributions with enhanced robustness to outliers. Additionally, I develop a proximal nonlinear conjugate gradient algorithm that accelerates convergence while maintaining numerical stability, along with rigorous statistical properties for the proposed framework.

AI Insights

• Conjugate Gradient, a family of optimization algorithms, solves linear systems and nonlinear problems efficiently.
• Descent Algorithms iteratively update estimates using gradient information to home in on minima or maxima.
• Elastic Net blends L1 and L2 penalties, striking a balance between sparsity and grouping effects.
• Tsallis entropy maximization extends Gaussian models to heavy‑tailed, outlier‑robust distributions.
• Proximal nonlinear conjugate gradient accelerates convergence while preserving numerical stability.
• Sparse recovery techniques thrive in high‑dimensional settings by exploiting underlying sparsity.
• Jeremy Watt’s “Machine Learning Refined” and Trefethen & Bau’s “Numerical Linear Algebra” are must‑reads for modern practitioners.

Abstract
Human-AI collaboration increasingly drives decision-making across industries, from medical diagnosis to content moderation. While AI systems promise efficiency gains by providing automated suggestions for human review, these workflows can trigger cognitive biases that degrade performance. We know little about the psychological factors that determine when these collaborations succeed or fail. We conducted a randomized experiment with 2,784 participants to examine how task design and individual characteristics shape human responses to AI-generated suggestions. Using a controlled annotation task, we manipulated three factors: AI suggestion quality in the first three instances, task burden through required corrections, and performance-based financial incentives. We collected demographics, attitudes toward AI, and behavioral data to assess four performance metrics: accuracy, correction activity, overcorrection, and undercorrection. Two patterns emerged that challenge conventional assumptions about human-AI collaboration. First, requiring corrections for flagged AI errors reduced engagement and increased the tendency to accept incorrect suggestions, demonstrating how cognitive shortcuts influence collaborative outcomes. Second, individual attitudes toward AI emerged as the strongest predictor of performance, surpassing demographic factors. Participants skeptical of AI detected errors more reliably and achieved higher accuracy, while those favorable toward automation exhibited dangerous overreliance on algorithmic suggestions. The findings reveal that successful human-AI collaboration depends not only on algorithmic performance but also on who reviews AI outputs and how review processes are structured. Effective human-AI collaborations require consideration of human psychology: selecting diverse evaluator samples, measuring attitudes, and designing workflows that counteract cognitive biases.

AI Insights

• Confirmation bias makes reviewers accept AI suggestions that confirm their beliefs, even if wrong.
• Intrinsic motivation dampens the effect of monetary incentives, reducing overreliance on algorithmic output.
• Simpler tasks amplify the influence of first impressions, making quality of AI suggestions more critical.
• Perceived fairness in pay can either curb or reinforce cognitive shortcuts during error correction.
• Transparent AI explanations improve error detection, offering a lever to counteract confirmation bias.
• Cognitive Illusions: A Handbook on Fallacies and Biases in Thinking, Judgement and Memory provides a taxonomy for these biases.
• Anchoring Bias Affects Mental Model Formation and User Reliance in Explainable AI Systems shows how early cues shape trust.

Abstract
Large language models (LLMs) distinguish themselves from previous technologies by functioning as collaborative "thought partners," capable of engaging more fluidly in natural language. As LLMs increasingly influence consequential decisions across diverse domains from healthcare to personal advice, the risk of overreliance - relying on LLMs beyond their capabilities - grows. This position paper argues that measuring and mitigating overreliance must become central to LLM research and deployment. First, we consolidate risks from overreliance at both the individual and societal levels, including high-stakes errors, governance challenges, and cognitive deskilling. Then, we explore LLM characteristics, system design features, and user cognitive biases that - together - raise serious and unique concerns about overreliance in practice. We also examine historical approaches for measuring overreliance, identifying three important gaps and proposing three promising directions to improve measurement. Finally, we propose mitigation strategies that the AI research community can pursue to ensure LLMs augment rather than undermine human capabilities.

AI Insights

• Explanations reduce overreliance, but they do not fully prevent decision errors.
• Uncertainty highlighting in code completions improves collaboration, yet requires careful design.
• Trust in ML models is complex; accuracy alone does not dictate user confidence.
• LLMs often fall short of human expectations, highlighting the need for capability audits.
• Papers 112, 121, 123, 128, 129, 131, 133 explore overreliance and mitigation.
• Human‑AI collaboration: a bidirectional loop where humans and models iteratively refine decisions.
• Uncertainty highlighting surfaces low‑confidence predictions to prompt human oversight.

Abstract
The deployment of capable AI agents raises fresh questions about safety, human-machine relationships and social coordination. We argue for greater engagement by scientists, scholars, engineers and policymakers with the implications of a world increasingly populated by AI agents. We explore key challenges that must be addressed to ensure that interactions between humans and agents, and among agents themselves, remain broadly beneficial.

Abstract
The challenge of aligning artificial intelligence (AI) with human values persists due to the abstract and often conflicting nature of moral principles and the opacity of existing approaches. This paper introduces CogniAlign, a multi-agent deliberation framework based on naturalistic moral realism, that grounds moral reasoning in survivability, defined across individual and collective dimensions, and operationalizes it through structured deliberations among discipline-specific scientist agents. Each agent, representing neuroscience, psychology, sociology, and evolutionary biology, provides arguments and rebuttals that are synthesized by an arbiter into transparent and empirically anchored judgments. We evaluate CogniAlign on classic and novel moral questions and compare its outputs against GPT-4o using a five-part ethical audit framework. Results show that CogniAlign consistently outperforms the baseline across more than sixty moral questions, with average performance gains of 16.2 points in analytic quality, 14.3 points in breadth, and 28.4 points in depth of explanation. In the Heinz dilemma, for example, CogniAlign achieved an overall score of 89.2 compared to GPT-4o's 69.2, demonstrating a decisive advantage in handling moral reasoning. By reducing black-box reasoning and avoiding deceptive alignment, CogniAlign highlights the potential of interdisciplinary deliberation as a scalable pathway for safe and transparent AI alignment.

Abstract
The impressive capabilities of deep learning models are often counterbalanced by their inherent opacity, commonly termed the "black box" problem, which impedes their widespread acceptance in high-trust domains. In response, the intersecting disciplines of interpretability and explainability, collectively falling under the Explainable AI (XAI) umbrella, have become focal points of research. Although these terms are frequently used as synonyms, they carry distinct conceptual weights. This document offers a comparative exploration of interpretability and explainability within the deep learning paradigm, carefully outlining their respective definitions, objectives, prevalent methodologies, and inherent difficulties. Through illustrative examinations of the MNIST digit classification task and IMDB sentiment analysis, we substantiate a key argument: interpretability generally pertains to a model's inherent capacity for human comprehension of its operational mechanisms (global understanding), whereas explainability is more commonly associated with post-hoc techniques designed to illuminate the basis for a model's individual predictions or behaviors (local explanations). For example, feature attribution methods can reveal why a specific MNIST image is recognized as a '7', and word-level importance can clarify an IMDB sentiment outcome. However, these local insights do not render the complex underlying model globally transparent. A clear grasp of this differentiation, as demonstrated by these standard datasets, is vital for fostering dependable and sound artificial intelligence.

Abstract
With the rapid advancement of large language models (LLMs), Multi-agent Systems (MAS) have achieved significant progress in various application scenarios. However, substantial challenges remain in designing versatile, robust, and efficient platforms for agent deployment. To address these limitations, we propose \textbf{LightAgent}, a lightweight yet powerful agentic framework, effectively resolving the trade-off between flexibility and simplicity found in existing frameworks. LightAgent integrates core functionalities such as Memory (mem0), Tools, and Tree of Thought (ToT), while maintaining an extremely lightweight structure. As a fully open-source solution, it seamlessly integrates with mainstream chat platforms, enabling developers to easily build self-learning agents. We have released LightAgent at \href{https://github.com/wxai-space/LightAgent}{https://github.com/wxai-space/LightAgent}

AI Insights

• LightAgent’s swarm design lets dozens of agents coordinate via one LightSwarm instance, boosting throughput.
• Each agent carries a distinct instruction set, enabling domain‑specific roles such as code synthesis or data retrieval.
• A built‑in text UI turns user prompts into executable code snippets, streamlining rapid prototyping.
• Tree‑of‑Thought logic lets agents iteratively refine plans, cutting hallucinations and improving accuracy.
• The lightweight core keeps memory usage under 200 MB on a single GPU while still supporting custom tool plugins.
• Advanced features can be daunting for beginners, and highly specialized tasks may still need manual tuning.
• LightAgent has been applied to robotics, finance, and healthcare, proving its versatility beyond chat‑bot demos.