IJCAI 2026 Papers

Neuro-symbolic systems are a promising paradigm for Hybrid Intelligence (HI), combining the complementary strengths of symbolic reasoning and machine learning in human-AI interactions. However, designing architectures that satisfy desired properties currently relies on informal reasoning, rather than a rigorous framework. Our research addresses this gap by developing a formal intermediate layer that links architecture design to system properties. Building on the visual pattern language of the Boxology for neuro-symbolic systems, we develop a mathematical formalization that is grounded in Category Theory, in order to enable proof-based guarantees of properties. The soundness of this formalism will be demonstrated through empirical case studies of existing Hybrid Intelligence systems. This work advances the theoretical foundations of neuro-symbolic AI by supporting principled design of HI systems.

In this era of Artifical Intelligence (AI), cyber threats are no longer static, as modern attackers are intelligent, adaptive, and capable of exploiting vulnerabilities faster than traditional defences can respond. My research investigates how attack and defence strategies co-evolve in complex, adversarial environments by integrating evolutionary game theory (EGT) with AI-driven multi-agent systems (MAS). This combined framework allows me to model strategic interactions between competing actors, analyse how defensive decisions influence overall security outcomes, and identify the conditions under which defences hold or break down. A particular focus of my work is the emergence of coordinated behaviours in multi-agent AI systems that can compromise system reliability in ways that existing safety approaches do not adequately address. Through theoretical modelling and controlled multi-agent experiments, my research proposes mechanisms that reduce attack success and strengthen defensive resilience. This work contributes to the understanding of adaptive cyber conflicts and the development of more robust and trust-worthy AI-enabled digital systems.

Equivariant graph neural networks (EGNNs) have revolutionized molecular property prediction by encoding 3D structures while preserving physical symmetry. However, existing EGNNs struggle to capture fine-grained geometric features needed for tasks like understanding protein interactions, and designing molecules that modulate them are central challenges in computational drug discovery. This thesis develops a progressive framework based on equivariant graph neural networks (EqGNNs) that respects the geometric symmetries of molecular structures. Our contributions progress along two axes: from discriminative to generative and from basic representations to physics-informed, language-conditioned generation. In this study, we propose novel methods for both discriminative (E(Q)AGNN-PPIS, GDEGAN) and generative tasks (PhysFlow). Together, these advance symmetry-aware geometric deep learning for drug discovery.

Federated Learning (FL) facilitates collaborative training of a global model whose performance is boosted by private data owned by distributed clients, without compromising data privacy. Yet the wide applicability of FL is hindered by entanglement of data distributions across different clients. This paper demonstrates for the first time that by disentangling data distributions FL can in principle achieve efficiencies comparable to those of distributed systems, requiring only one round of communication. To this end, we propose a novel FedDistr algorithm, which employs stable diffusion models to decouple and recover data distributions. Empirical results on the CIFAR100 and DomainNet datasets show that FedDistr significantly enhances model utility and efficiency in both disentangled and near-disentangled scenarios while ensuring privacy, outperforming traditional federated learning methods.

Fairness in machine learning (ML) has garnered significant attention. However, current research has mainly concentrated on the distributive fairness of ML models, with limited focus on another dimension of fairness, i.e., procedural fairness. In this paper, we first define the procedural fairness of ML models by drawing from the established understanding of procedural fairness in philosophy and psychology fields, and then give formal definitions of individual and group procedural fairness. Based on the proposed definition, we further propose a novel metric to evaluate the group procedural fairness of ML models, called GPFFAE, which utilizes a widely used explainable artificial intelligence technique, namely feature attribution explanation (FAE), to capture the decision process of ML models. We validate the effectiveness of GPFFAE on a synthetic dataset and eight real-world datasets. Our experimental studies have revealed the relationship between procedural and distributive fairness of ML models. After validating the proposed metric for assessing the procedural fairness of ML models, we then propose a method for identifying the features that lead to the procedural unfairness of the model and propose two methods to improve procedural fairness based on the identified unfair features. Our experimental results demonstrate that we can accurately identify the features that lead to procedural unfairness in the ML model, and both of our proposed methods can significantly improve procedural fairness while also improving distributive fairness, with a slight sacrifice on the model performance

Background: Trustworthy AI serves as a foundational pillar for two major AI ethics conferences: AIES and FAccT. However, current research often adopts techno-centric approaches, focusing primarily on technical attributes such as reliability, robustness, and fairness, while overlooking the sociotechnical dimensions critical to understanding AI trustworthiness in real-world contexts. Objectives: This scoping review aims to examine how the AIES and FAccT communities conceptualize, measure, and validate AI trustworthiness, identifying major gaps and opportunities for advancing a holistic understanding of trustworthy AI systems. Methods: We conduct a scoping review of AIES and FAccT conference proceedings to date, systematically analyzing how trustworthiness is defined, operationalized, and applied across different research domains. Our analysis focuses on conceptualization approaches, measurement methods, verification and validation techniques, application areas, and underlying values. Results: While significant progress has been made in defining technical attributes such as transparency, accountability, and robustness, our findings reveal critical gaps. Current research often predominantly emphasizes technical precision at the expense of social and ethical considerations. The sociotechnical nature of AI systems remains less explored and trustworthiness emerges as a contested concept shaped by those with the power to define it. Conclusions: An interdisciplinary approach combining technical rigor with social, cultural, and institutional considerations is essential for advancing trustworthy AI. We propose actionable measures for the AI ethics community to adopt holistic frameworks that genuinely address the complex interplay between AI systems and society, ultimately promoting responsible technological development that benefits all stakeholders.

With the impressive progress of deep learning, applications relying on machine learning are increasingly being integrated into daily life. However, most deep learning models have an opaque, oracle-like nature making it difficult to interpret and understand their decisions. This problem led to the development of the field known as eXplainable Artificial Intelligence (XAI). One method in this field known as Projective Simulation (PS) models a chain-of-thought as a random walk of a particle on a graph with vertices that have concepts attached to them. While this description has various benefits, including the possibility of quantization, it cannot be naturally used to model thoughts that combine several concepts simultaneously. To overcome this limitation, we introduce Multi-Excitation Projective Simulation (mePS), a generalization that considers a chain-of-thought to be a random walk of several particles on a hypergraph. A definition for a dynamic hypergraph is put forward to describe the agent's training history along with applications to AI and hypergraph visualization. An inductive bias inspired by the remarkably successful few-body interaction models used in quantum many-body physics is formalized for our classical mePS framework and employed to tackle the exponential complexity associated with naive implementations of hypergraphs. We prove that our inductive bias reduces the complexity from exponential to polynomial, with the exponent representing the cutoff on how many particles can interact. We numerically apply our method to two toy environments and a more complex scenario modelling the diagnosis of a broken computer. These environments demonstrate the resource savings provided by an appropriate choice of inductive bias, as well as showcasing aspects of interpretability. A quantum model for mePS is also briefly outlined and some future directions for it are discussed.

We study the fair and truthful allocation of m divisible public items among n agents, each with distinct preferences for the items. To aggregate agents' preferences fairly, we focus on finding a core solution. For divisible items, a core solution always exists and can be calculated by maximizing the Nash welfare objective. However, such a solution is easily manipulated; agents might have incentives to misreport their preferences. To mitigate this, the current state-of-the-art finds an approximate core solution with high probability while ensuring approximate truthfulness. However, this approach has two main limitations. First, due to several approximations, the approximation error in the core could grow with n, resulting in a non-asymptotic core solution. This limitation is particularly significant as public-good allocation mechanisms are frequently applied in scenarios involving a large number of agents, such as the allocation of public tax funds for municipal projects. Second, implementing the current approach for practical applications proves to be a highly nontrivial task. To address these limitations, we introduce PPGA, a (differentially) Private Public-Good Allocation algorithm, and show that it attains asymptotic truthfulness and finds an asymptotic core solution with high probability. Additionally, to demonstrate the practical applicability of our algorithm, we implement PPGA and empirically study its properties using municipal participatory budgeting data.

Recently, we are witnessing an increase in computation power and memory, leading to strong AI algorithms becoming applicable in areas affecting our daily lives. We focus on AI planning solutions for complex, real-life decision-making problems under uncertainty, such as autonomous driving. Human trust in such AI-based systems is essential for their acceptance and market penetration. Moreover, users need to establish appropriate levels of trust to benefit the most from these systems. Previous studies have motivated this work, showing that users can benefit from receiving (handcrafted) information about the reasoning of a stochastic AI planner, for example, controlling automated driving maneuvers. Our solution to automating these hand-crafted notifications with explainable AI algorithms, XAI, includes studying: (1) what explanations can be generated from an AI planning system, applied to a real-world problem, in real-time? What is that content that can be processed from a planner's reasoning that can help users understand and trust the system controlling a behavior they are experiencing? (2) when can this information be displayed? and (3) how shall we display this information to an end user? The value of these computed XAI notifications has been assessed through an online user study with 800 participants, experiencing simulated automated driving scenarios. Our results show that real time XAI notifications decrease significantly subjective misunderstanding of participants compared to those that received only a dynamic HMI display. Also, our XAI solution significantly increases the level of understanding of participants with prior ADAS experience and of participants that lack such experience but have non-negative prior trust to ADAS features. The level of trust significantly increases when XAI was provided to a more restricted set of the participants, including those over 60 years old, with prior ADAS experience and non-negative prior trust attitude to automated features.

Graph Neural Networks (GNNs) have demonstrated remarkable proficiency in modeling data with graph structures, yet recent research reveals their susceptibility to adversarial attacks. Traditional attack methodologies, which rely on manipulating the original graph or adding links to artificially created nodes, often prove impractical in real-world settings. This paper introduces a novel adversarial scenario involving the injection of an isolated subgraph to deceive both the link recommender and the node classifier within a GNN system. Specifically, the link recommender is mislead to propose links between targeted victim nodes and the subgraph, encouraging users to unintentionally establish connections and that would degrade the node classification accuracy, thereby facilitating a successful attack. To address this, we present the LiSA framework, which employs a dual surrogate model and bi-level optimization to simultaneously meet two adversarial objectives. Extensive experiments on real-world datasets demonstrate the effectiveness of our method.

Deep neural networks are vulnerable to adversarial examples, posing serious security risks. While GAN-based attacks such as AdvGAN enable fast adversarial example generation, they typically operate in a single step, limiting attack effectiveness and transferability. In this work, we propose PAR-AdvGAN, a progressive auto-regressive GAN framework that iteratively refines adversarial perturbations. By conditioning each iteration on the previous output, PAR-AdvGAN produces stronger, more transferable attacks while maintaining low visual distortion and high generation speed. Experiments on multiple models and datasets demonstrate that PAR-AdvGAN significantly outperforms baseline methods in attack success rate, achieving up to 335.5 frames per second, highlighting its practical efficiency for black-box attack scenarios.

Deep neural networks often rely on spurious correlations, i.e., superficial associations between non-causal features and prediction targets. While yielding high overall accuracy during training, such reliance degrades generalization on minority groups where these correlations break down. Existing methods mitigate this issue by using external group annotations or auxiliary deterministic models, but group annotations are costly to obtain and deterministic auxiliaries may fail to capture the full spectrum of biases learned by the model. We propose Evidential Alignment, a framework that leverages uncertainty quantification to identify and suppress spurious correlations without requiring group annotations. By transforming a biased ERM model from first-order to second-order risk minimization with a Dirichlet distribution and applying an evidential calibration step that reweighs samples by their epistemic uncertainty, Evidential Alignment debiases the model while preserving core features. We provide theoretical guarantees and demonstrate strong worst-group accuracy across diverse architectures and modalities.

Large Language Models are typically trained with next-turn rewards, limiting their ability to optimize for long-term interaction. As a result, they often respond passively to ambiguous or open-ended user requests, failing to help users reach their ultimate intents and leading to inefficient conversations. To address these limitations, we introduce CollabLLM, a novel and general training framework that enhances multiturn human-LLM collaboration. Its key innovation is a collaborative simulation that estimates the long-term contribution of responses using Multiturn-aware Rewards. By reinforcement fine-tuning these rewards, CollabLLM goes beyond responding to user requests and actively uncovers user intent and offers insightful suggestions--a key step toward more human-centered AI. We also devise a multiturn interaction benchmark with three challenging tasks such as document creation. COLLABLLM significantly outperforms our baselines with averages of 18.5% higher task performance and 44.3% improved interactivity by LLM judges. Finally, we conduct a large user study with 201 judges, where CollabLLM increases user satisfaction by 17.6% and reduces user spent time by 10.4%.

Large Language Models (LLMs) are often used in autonomous decision-making, where they have to sample options from vast action spaces. Here we present a summary of the work studying the heuristics that guide this sampling process and show it resembles that of human decision-making: comprising a descriptive component (reflecting statistical norm) and a prescriptive component (implicit ideal encoded in the LLM). We show that the deviation of a sample from the statistical norm towards a prescriptive component consistently appears in concepts across diverse real-world domains. To further illustrate the theory, we demonstrate that concept prototypes in LLMs are affected by prescriptive norms, similar to the concept of normality in humans. This finding has implications in LLM based agentic systems, explaining their exploration behavior and biases in decision making.

Counterfactual explanations have become an important paradigm for improving the transparency of machine learning models by showing how small input changes can alter model outputs. While substantial progress has been made in generating such explanations, their evaluation remains insufficiently standardized, particularly for systems that produce ranked outputs rather than single-label predictions. Existing evaluation protocols in recommender systems commonly focus only on whether the top-1 recommendation changes after perturbation. We argue that this practice can lead to inconsistent and misleading conclusions, as the relative ranking of explanation methods may vary with changes in the quality of the underlying model. In this work, we revisit the evaluation of counterfactual explanations from a ranking perspective. We propose extending top-1 evaluation to list-wise top-k protocols that assess explanation effectiveness across multiple highly ranked outputs. Through experiments on multiple datasets, recommender architectures, and explanation methods, we show that top-k evaluation substantially improves consistency and yields more reliable comparisons between competing explainers. Our findings highlight a broader methodological lesson for explainable AI: when models return ranked results, explanation quality should be assessed using ranking-aware evaluation protocols rather than top-1 criteria alone.

Driving progress of AI models and agents requires comparing their performance on standardized benchmarks; for general agents, individual performances must be aggregated across a potentially wide variety of different tasks. In this extended abstract, we describe a ranking scheme inspired by social choice frameworks, called Soft Condorcet Optimization (SCO), to compute the optimal ranking of agents: the one that makes the fewest mistakes in predicting the agent comparisons in the evaluation data. This optimal ranking is the maximum likelihood estimate when evaluation data (which we view as votes) are interpreted as noisy samples from a ground truth ranking, a solution to Condorcet's original voting system criteria. SCO ratings are maximal for Condorcet winners when they exist, which we show is not necessarily true for the classical rating system Elo. In practice, SCO serves as an accurate approximation to the Kemeny-Young voting method, excels in the sparse data regime, and provides the best approximation to the optimal ranking compared to every baseline on a Diplomacy player ranking problem with 31,094 games and 52,958 agents.

A fundamental task in multi-agent systems is to match n agents to n alternatives (e.g., resources or tasks). This is often done by eliciting agents' ordinal rankings over the alternatives rather than their exact numerical utilities. While this simplifies elicitation, the incomplete information leads to inefficiency, captured by a worst-case measure called distortion. Recent work shows that making just a few cardinal utility queries per agent can significantly improve the distortion, and in particular that O(√n) distortion is achievable with two queries per agent. We generalize this result by achieving O(n^(1/λ)) distortion with λ queries per agent, for any constant λ, which is optimal up to a constant factor given a previously known lower bound. We extend this finding to the general social choice problem of selecting one of m alternatives based on n agents' preferences, achieving O(min(n, m)^(1/λ)) distortion with λ queries per agent.

Recommenders are significantly shaping online information consumption. While effective at personalizing content, these systems increasingly face criticism for propagating irrelevant, unwanted, and even harmful recommendations. Such content degrades user satisfaction and contributes to significant societal issues, including misinformation, radicalization, and erosion of user trust. Although platforms offer mechanisms to mitigate exposure to undesired content, these mechanisms are often insufficiently effective and slow to adapt to users' feedback. This paper introduces an intuitive, model-agnostic, and distribution-free method that uses conformal risk control to provably bound unwanted content in personalized recommendations by leveraging simple binary feedback on items. We also address a limitation of traditional conformal risk control approaches, i.e., the fact that the recommender can provide a smaller set of recommended items, by leveraging implicit feedback on consumed items to expand the recommendation set while ensuring robust risk mitigation. Our experimental evaluation on data coming from a popular online video-sharing platform demonstrates that our approach ensures an effective and controllable reduction of unwanted recommendations with minimal effort.

In many real-world scenarios, interactions happen in a group-wise manner with multiple entities, and therefore, hypergraphs are a suitable tool to accurately represent such interactions. Various generative models have been proposed to explore fundamental mechanisms underlying hyperedge formation. However, most existing models do not account for node attributes, which can play a significant role in hyperedge formation. As a result, they fail to capture the interactions between structure and node attributes. To address this issue, we propose NoAH, a stochastic hypergraph generative model for attributed hypergraphs. NoAH utilizes the core–fringe node hierarchy to model hyperedge formation as a series of node attachments and determines attachment probabilities based on node attributes. We further introduce NoAHFit, a parameter learning procedure for fitting NoAH to a given real-world hypergraph. Through experiments, we show that NoAH with NoAHFit more accurately reproduces the structure–attribute interplay than baseline hypergraph generative models. Our code, supplementary materials, and datasets are provided at https://github.com/jaewan01/NoAH.