Research Papers

A generative adversarial network (GAN) has been a representative backbone model in generative artificial intelligence (AI) because of its powerful performance in capturing intricate data-generating processes. However, the GAN training is well-known for its notorious training instability, usually characterized by the occurrence of mode collapse. Through the lens of gradients' variance, this work particularly analyzes the training instability and inefficiency in the presence of mode collapse by linking it to multimodality in the target distribution. To ease the raised training issues from severe multimodality, we introduce a novel GAN training framework that leverages a series of tempered distributions produced via convex interpolation. With our newly developed GAN objective function, the generator can learn all the tempered distributions simultaneously, conceptually resonating with the parallel tempering in statistics. Our simulation studies demonstrate the superiority of our approach over existing popular training strategies in both image and tabular data synthesis. We theoretically analyze that such significant improvement can arise from reducing the variance of gradient estimates by using the tempered distributions. Finally, we further develop a variant of the proposed framework aimed at generating fair synthetic data which is one of the growing interests in the field of trustworthy AI.

High-impact and potentially dangerous capabilities can and should be broken down into early warning shots long before reaching red lines. Each of these early warning shots should correspond to a precursory capability. Each precursory capability sits on a spectrum indicating its proximity to a final high-impact capability, corresponding to a red line. To meaningfully detect and track capability progress, we propose a taxonomy of dangerous capability zones (a zoning taxonomy) tied to a staggered information exchange framework that enables relevant bodies to take action accordingly. In the Frontier AI Safety Commitments, signatories commit to sharing more detailed information with trusted actors, including an appointed body, as appropriate (Commitment VII). Building on our zoning taxonomy, this paper makes four recommendations for specifying information sharing as detailed in Commitment VII. (1) Precursory capabilities should be shared as soon as they become known through internal evaluations before deployment. (2) AI Safety Institutes (AISIs) should be the trusted actors appointed to receive and coordinate information on precursory components. (3) AISIs should establish adequate information protection infrastructure and guarantee increased information security as precursory capabilities move through the zones and towards red lines, including, if necessary, by classifying the information on precursory capabilities or marking it as controlled. (4) High-impact capability progress in one geographical region may translate to risk in other regions and necessitates more comprehensive risk assessment internationally. As such, AISIs should exchange information on precursory capabilities with other AISIs, relying on the existing frameworks on international classified exchanges and applying lessons learned from other regulated high-risk sectors.

Model interpretability is a key challenge that has yet to align with the advancements observed in contemporary state-of-the-art deep learning models. In particular, deep learning aided vision tasks require interpretability, in order for their adoption in more specialized domains such as medical imaging. Although the field of explainable AI (XAI) developed methods for interpreting vision models along with early convolutional neural networks, recent XAI research has mainly focused on assigning attributes via saliency maps. As such, these methods are restricted to providing explanations at a sample level, and many explainability methods suffer from low adaptability across a wide range of vision models. In our work, we re-think vision-model explainability from a novel perspective, to probe the general input structure that a model has learnt during its training. To this end, we ask the question: "How would a vision model fill-in a masked-image". Experiments on standard vision datasets and pre-trained models reveal consistent patterns, and could be intergrated as an additional model-agnostic explainability tool in modern machine-learning platforms. The code will be available at \url{https://github.com/BoTZ-TND/FillingTheBlanks.git}

The downstream use cases, benefits, and risks of AI models depend significantly on what sort of access is provided to the model, and who it is provided to. Though existing safety frameworks and AI developer usage policies recognise that the risk posed by a given model depends on the level of access provided to a given audience, the procedures they use to make decisions about model access are ad hoc, opaque, and lacking in empirical substantiation. This paper consequently proposes that frontier AI companies build on existing safety frameworks by outlining transparent procedures for making decisions about model access, which we term Responsible Access Policies (RAPs). We recommend that, at a minimum, RAPs should include the following: i) processes for empirically evaluating model capabilities given different styles of access, ii) processes for assessing the risk profiles of different categories of user, and iii) clear and robust pre-commitments regarding when to grant or revoke specific types of access for particular groups under specified conditions.

In explanations, explainers have mental representations of explainees' developing knowledge and shifting interests regarding the explanandum. These mental representations are dynamic in nature and develop over time, thereby enabling explainers to react to explainees' needs by adapting and customizing the explanation. XAI should be able to react to explainees' needs in a similar manner. Therefore, a component that incorporates aspects of explainers' mental representations of explainees is required. In this study, we took first steps by investigating explainers' mental representations in everyday explanations of technological artifacts. According to the dual nature theory, technological artifacts require explanations with two distinct perspectives, namely observable and measurable features addressing "Architecture" or interpretable aspects addressing "Relevance". We conducted extended semi structured pre-, post- and video recall-interviews with explainers (N=9) in the context of an explanation. The transcribed interviews were analyzed utilizing qualitative content analysis. The explainers' answers regarding the explainees' knowledge and interests with regard to the technological artifact emphasized the vagueness of early assumptions of explainers toward strong beliefs in the course of explanations. The assumed knowledge of explainees in the beginning is centered around Architecture and develops toward knowledge with regard to both Architecture and Relevance. In contrast, explainers assumed higher interests in Relevance in the beginning to interests regarding both Architecture and Relevance in the further course of explanations. Further, explainers often finished the explanation despite their perception that explainees still had gaps in knowledge. These findings are transferred into practical implications relevant for user models for adaptive explainable systems.

This study presents DhondtXAI as a SHAP-independent, D'Hondt-based attribution framework for tabular XAI. Instead of model-native feature importance or SHAP values, DhondtXAI computes background-interventional removal effects, separates positive and negative evidence, forms optional feature alliances, applies optional thresholds, allocates seats via the D'Hondt rule, and projects onto the local model-output difference. Completeness is preserved by construction, with the projection residual ratio reported as a diagnostic. The method is evaluated on synthetic additive and interaction tests, correlated-feature perturbations, operator and apportionment ablations, projection-mode comparisons, logit-scale checks, repeated split validation, paired deletion tests, and two healthcare datasets: Wisconsin Diagnostic Breast Cancer (CatBoost) and early-stage diabetes risk prediction (XGBoost). SHAP serves only as an external comparator with aligned settings. In additive synthetics, DhondtXAI exactly recovers ground-truth rankings; in multiplicative interactions, alliances reduce the mean projection residual from 0.2527 to 0.0001. On WDBC and diabetes data, it shows high agreement with SHAP (Spearman rho = 0.9273 and 0.9353), supported by further signed, top-k, magnitude, deletion, and sensitivity analyses. Results position DhondtXAI as a complementary proportional, alliance-aware, and threshold-aware tabular XAI method, not a replacement for SHAP or LIME.

Understanding node representations in graph-based models is crucial for uncovering biases ,diagnosing errors, and building trust in model decisions. However, previous work on explainable AI for node representations has primarily emphasized explanations (reasons for model predictions) rather than interpretations (mapping representations to understandable concepts). Furthermore, the limited research that focuses on interpretation lacks validation, and thus the reliability of such methods is unclear. We address this gap by proposing a novel interpretation method-Node Coherence Rate for Representation Interpretation (NCI)-which quantifies how well different node relations are captured in node representations. We also propose a novel method (IME) to evaluate the accuracy of different interpretation methods. Our experimental results demonstrate that NCI reduces the error of the previous best approach by an average of 39%. We then apply NCI to derive insights about the node representations produced by several graph-based methods and assess their quality in unsupervised settings.

There have been works discussing the adoption of a human rights framework for responsible AI, emphasizing various rights such as the right to contribute to scientific advancements. Yet, to the best of our knowledge, this is the first attempt to take this framework with special focus on computer vision and documenting human rights violations in its community. This work summarizes such incidents accompanied with evidence from the lens of a female African Muslim Hijabi researcher. While previous works resorted to qualitative surveys that gather opinions from various researchers in the field, this work argues that a single documented violation is sufficient to warrant attention regardless of the stature of this researcher. Incidents documented in this work include silence on Genocides that are occurring while promoting the governments contributing to it, a broken reviewing system and corruption in the faculty support systems. This work discusses that demonizing individuals for discrimination based on gender, ethnicity, creed or reprisal has been a successful tool for exclusion with documented evidence from a single case. We argue that human rights are guaranteed for every single individual even the ones that might be labelled as devils in the community for whichever reasons to dismantle such a tool from its roots.

We present a theoretical framework assessing the economic implications of bias in AI-powered emergency response systems. Integrating health economics, welfare economics, and artificial intelligence, we analyze how algorithmic bias affects resource allocation, health outcomes, and social welfare. By incorporating a bias function into health production and social welfare models, we quantify its impact on demographic groups, showing that bias leads to suboptimal resource distribution, increased costs, and welfare losses. The framework highlights efficiency-equity trade-offs and provides economic interpretations. We propose mitigation strategies, including fairness-constrained optimization, algorithmic adjustments, and policy interventions. Our findings offer insights for policymakers, emergency service providers, and technology developers, emphasizing the need for AI systems that are efficient and equitable. By addressing the economic consequences of biased AI, this study contributes to policies and technologies promoting fairness, efficiency, and social welfare in emergency response services.

Federated learning is a prominent distributed learning paradigm that incorporates collaboration among diverse clients, promotes data locality, and thus ensures privacy. These clients have their own technological, cultural, and other biases in the process of data generation. However, the present standard often ignores this bias/heterogeneity, perpetuating bias against certain groups rather than mitigating it. In response to this concern, we propose an equitable clustering-based framework where the clients are categorized/clustered based on how similar they are to each other. We propose a unique way to construct the similarity matrix that uses activation vectors. Furthermore, we propose a client weighing mechanism to ensure that each cluster receives equal importance and establish $O(1/\sqrt{K})$ rate of convergence to reach an $ε-$stationary solution. We assess the effectiveness of our proposed strategy against common baselines, demonstrating its efficacy in terms of reducing the bias existing amongst various client clusters and consequently ameliorating algorithmic bias against specific groups.

As AI systems increasingly integrate into critical societal sectors, the demand for robust privacy-preserving methods has escalated. This paper introduces Data Obfuscation through Latent Space Projection (LSP), a novel technique aimed at enhancing AI governance and ensuring Responsible AI compliance. LSP uses machine learning to project sensitive data into a latent space, effectively obfuscating it while preserving essential features for model training and inference. Unlike traditional privacy methods like differential privacy or homomorphic encryption, LSP transforms data into an abstract, lower-dimensional form, achieving a delicate balance between data utility and privacy. Leveraging autoencoders and adversarial training, LSP separates sensitive from non-sensitive information, allowing for precise control over privacy-utility trade-offs. We validate LSP's effectiveness through experiments on benchmark datasets and two real-world case studies: healthcare cancer diagnosis and financial fraud analysis. Our results show LSP achieves high performance (98.7% accuracy in image classification) while providing strong privacy (97.3% protection against sensitive attribute inference), outperforming traditional anonymization and privacy-preserving methods. The paper also examines LSP's alignment with global AI governance frameworks, such as GDPR, CCPA, and HIPAA, highlighting its contribution to fairness, transparency, and accountability. By embedding privacy within the machine learning pipeline, LSP offers a promising approach to developing AI systems that respect privacy while delivering valuable insights. We conclude by discussing future research directions, including theoretical privacy guarantees, integration with federated learning, and enhancing latent space interpretability, positioning LSP as a critical tool for ethical AI advancement.

Providing human-understandable insights into the inner workings of neural networks is an important step toward achieving more explainable and trustworthy AI. Existing approaches to such mechanistic interpretability typically require substantial prior knowledge and manual effort, with strategies tailored to specific tasks. In this work, we take a step toward automating the understanding of the network by investigating the existence of distinct sub-networks. Specifically, we explore a novel automated and task-agnostic approach based on the notion of functionally similar representations within neural networks to identify similar and dissimilar layers, revealing potential sub-networks. We achieve this by proposing, for the first time to our knowledge, the use of Gromov-Wasserstein distance, which overcomes challenges posed by varying distributions and dimensionalities across intermediate representations, issues that complicate direct layer to layer comparisons. On algebraic, language, and vision tasks, we observe the emergence of sub-groups within neural network layers corresponding to functional abstractions. Through downstream applications of model compression and fine-tuning, we show the proposed approach offers meaningful insights into the behavior of neural networks with minimal human and computational cost.