IJCAI 2026 Papers

Most of the widely used estimators of the average treatment effect (ATE) in causal inference rely on the assumptions of unconfoundedness and overlap. Unconfoundedness requires that the observed covariates account for all correlations between the outcome and treatment. Overlap requires the existence of randomness in treatment decisions for all individuals. Nevertheless, many types of studies frequently violate unconfoundedness or overlap; for instance, observational studies with deterministic treatment decisions, popularly known as Regression Discontinuity designs, violate overlap. In this paper, we initiate the study of general conditions that enable the identification of the average treatment effect, extending beyond unconfoundedness and overlap. In particular, following the paradigm of statistical learning theory, we provide an interpretable condition that is sufficient and necessary for the identification of ATE. Moreover, this condition can be used to characterize other treatment effects, such as the average treatment effect on the treated (ATT), as well. To illustrate the utility of our condition, we present several well-studied scenarios where our condition is satisfied and, hence, we prove that ATE can be identified in regimes that prior works could not capture. For example, under mild assumptions on the data distributions, this holds for the models proposed by Tan (2006) and Rosenbaum (2002), and the Regression Discontinuity design model introduced by Thistlethwaite and Campbell (1960). For each of these scenarios, we also show that, under natural additional assumptions, ATE can be estimated from finite samples. We believe these findings open new avenues for bridging learning-theoretic insights and causal inference methodologies, particularly in observational studies with complex treatment mechanisms.

Letting AI agents interact in multi-agent settings introduces significant complexity in predicting and interpreting their collective behavior, with profound implications for trustworthy AI adoption in research and society. We present FAIRGAME (Framework for AI Agents Bias Recognition using Game Theory), an open-source framework that simulates game-theoretic scenarios with LLM-based agents to systematically uncover biases arising from model choice, language, agent personality, and more. Applied to the Prisoner’s Dilemma and Battle of the Sexes across four LLMs and five human languages, FAIRGAME reveals inconsistencies across LLM models and consistent deviations from game-theoretic predictions; it also quantifies LLM-specific behavioral tendencies through a novel scoring system. Our results show that LLMs draw on prior world knowledge beyond payoff matrices, and that language and personality significantly shape strategic outcomes, supporting the use of reproducible and controlled simulation pipelines to predict the interacting behavior of LLM agents.

We establish a correspondence between (fragments of) TEL^o, a temporal extension of the EL description logic with the LTL operator O^k, and some specific kinds of formal grammars, in particular, conjunctive grammars (context-free grammars equipped with the operation of intersection). This connection implies that TEL^o does not possess the property of ultimate periodicity of models, and further leads to undecidability of query answering in TEL^o, closing a question left open since the introduction of TEL^o. Moreover, it also allows to establish decidability of query answering for some new interesting fragments of TEL^o, and to reuse for this purpose existing tools and algorithms for conjunctive grammars.

We present an industrial case on workforce allocation and scheduling in the aircraft manufacturing industry, where available teams need to be assigned to logistical operations. This application presents several challenges, such as the scale of the problem, the need for fair workload distribution, and the need for methods to mitigate unforeseen disruptions due to technical malfunctions or incompatible weather conditions. We compare different Constraint Programming (CP) models for the allocation and scheduling problems, with extra focus on modeling the workload balancing component. Additionally, we investigate automatic rescheduling methods for restoring feasibility after a disruption invalidates the precomputed schedule. Our results show that by using appropriate modeling techniques, the problem can be solved in reasonable time, thereby producing fair schedules. Additionally, we show how invalidated schedules can be restored efficiently to help human operators in resolving disruptions to the schedule.

Partially observable Markov decision processes (POMDPs) form a prominent model for uncertainty in sequential decision making. We are interested in constructing algorithms with theoretical guarantees to determine whether the agent has a strategy ensuring a given specification with probability 1. This well-studied problem is known to be undecidable already for very simple omega-regular objectives, because of the difficulty of reasoning on uncertain events. We introduce a revelation mechanism which restricts information loss by requiring that, almost surely, the agent has eventually full information about the current state. Our main technical results are to construct exact algorithms for two classes of POMDPs called weakly and strongly revealing. Importantly, the decidable cases reduce to the analysis of a finite belief-support Markov decision process. This yields a conceptually simple and exact algorithm for a large class of POMDPs.

The Multi-objective Shortest Path (MOSP) problem is a classic network optimization problem that aims to find all Pareto-optimal paths between two points in a graph with multiple edge costs. Recent studies on multi-objective search with A* (MOA*) have demonstrated strong performance on challenging MOSP instances. This paper presents a novel search framework that enables efficient parallelization of MOA* through different objective orderings and a unique upper-bounding strategy that, in certain cases, allows the problem dimensionality to be reduced to one. Results demonstrate that the proposed framework can significantly improve the performance of MOA*, with speedups increasing proportionally to the number of objectives.

Learned agents that control traffic signals or call software tools are usually trained in simulators or fixed benchmarks, yet must act in worlds that differ. Many reliability concerns (robustness, safety, alignment, etc.) become one problem once agents are written as Markov decision processes (MDPs): a gap between the world an agent was built for and the world it acts in. The sim-to-real gap decomposes along state, observation, action, transition, and reward; multi-agent deployment further adds topology, population, and task gaps as neighbors learn and workloads scale. Using reinforcement learning (RL) for traffic control and large language model (LLM) agents for tool use and mobile GUIs as running examples, this paper traces how mismatches appear on each channel and how mitigations from one community (domain randomization, grounded action transformation, perturbation benchmarks) transfer to the other. Because the gap rarely closes, reliable deployment needs selective human oversight; uncertainty quantification can flag when to escalate, localize errors at the answer, reasoning, step, or agent level, and support sim-to-real transfer in RL. This paper opens an agenda for a shared MDP vocabulary, uncertainty-aware oversight, and scaling human-AI collaboration across volume, complexity, and expertise.

Modern artificial intelligence has made rapid progress by scaling data, models, and computation. Yet scale alone does not solve one basic problem. Many intelligent systems must learn, adapt, and act when direct experience is scarce, costly, noisy, or changing. A scientist may have only a few labeled molecules. A recommender system may see only sparse interactions for a new user or item. A language-model agent may need to align with a user's preference from a short interaction history. These settings differ, but they share the same question: how can AI extract more learning signal from less experience? My research studies this question through data-efficient generalization. This article summarizes a research path from few-shot learning to meta-learning, in-context learning, and data-efficient agentic systems. The central theme is that limited supervision can be made useful by the right priors, the right adaptation mechanism, and the right way to reuse experience. I also discuss two regimes where data efficiency is not only desirable but necessary: scientific scarcity and efficient data utilization.

Human reasoning relies on abstraction and generalization, in order to make decisions flexible under changing conditions while ignoring irrelevant details and focusing on the essence. Developing AI systems with such abilities, while ensuring transparency and explainability on the reasoning behind the made decision, remains a central challenge. Symbolic AI provides transparent knowledge representations and formal reasoning guarantees, yet lacks principled mechanisms for abstracting away irrelevant details while preserving the information required for reasoning, explainability, and generalization. This paper presents an overview of my research towards addressing this challenge by developing formal and computational methods for abstraction in Answer Set Programming, one of the core formalisms in symbolic AI, and related logic-based frameworks. The contributions span from foundational abstraction techniques that simplify reasoning representations while preserving essential solution properties, to investigations on how such abstractions can both improve computational reasoning and support human understanding of AI decision processes.

The rapid advancement of Artificial Intelligence (AI) has accelerated the development of personalised healthcare. However, the clinical adoption of deep learning remains constrained by a persistent “trust gap” surrounding model transparency, security, and data privacy. This talk presents a research vision for building trustworthy human-centred AI in computer audition and biosignal processing by integrating signal processing, machine learning, and healthcare. This vision is structured around three interconnected pillars: Sensing, Understanding, and Empowerment. The Sensing pillar focuses on enabling machines to perceive clinically relevant information from biosignals, transforming body sounds and physiological signals into non-invasive, accessible, and cost-effective windows into human health and well-being. The Understanding pillar addresses the black-box nature of modern AI systems, with the goal of ensuring transparency, efficiency, robustness, and security in clinical deployment. This is achieved through the development of explainable AI methods, knowledge distillation techniques, and defences against adversarial attacks, fostering AI systems that clinicians can trust and interpret. The Empowerment pillar seeks to restore natural communication for individuals with speech impairments, such as laryngectomy patients. By developing Silent Speech Interfaces (SSIs) that translate facial muscle activity (EMG) directly into audible speech, this research advances speech intelligibility, naturalness, real-time causal architectures, and multi-speaker speech synthesis. Collectively, these research directions aim to establish a foundation of trustworthy, human-centred AI that is not only accurate and secure, but also empathetic, accessible, and impactful, ultimately enhancing healthcare, communication, and quality of life.

Many real-world problems can be modeled as cooperative multi-agent systems (MAS), such as fleet management, industrial operations, and communication networks, where multiple agents collaborate to optimize a shared objective. Optimizing cooperative MAS is difficult due to the combinatorial nature of joint actions and environmental factors. Thus, many practical multi-agent optimization approaches specialize in particular problem classes to exploit structural properties for effective and efficient optimization. Unfortunately, such specializations can lead to complex and inflexible methods that cannot be seamlessly combined or transferred to novel domains without substantial engineering effort. In this paper, we advocate an approach towards streamlined learning and search for multi-agent optimization. Focusing on multi-agent path finding as an exemplary problem, we propose to simplify two popular approaches to MAPF, namely multi-agent reinforcement learning and adaptive search. Through these simplifications, we aim to enable seamless combination and transferability of our methods without substantial engineering.

Deploying AI in real-world intelligent transportation systems (ITS) remains hard: traffic data is noisy and incomplete, learned policies rarely transfer to new cities, and black-box models cannot earn the trust of operators and public officials. Drawing on years of joint experience in ITS industry and academia (including traffic signal control systems deployed across multiple cities and AI for one of the world's busiest metro networks), this paper presents TrafficPDE: a practitioner's guidebook organized around the Perception-Decision-Explanation (PDE) triangle. Perception tackles how to model traffic data reliably despite noise, missing sensors, and unobserved locations. Decision develops generalizable reinforcement learning agents that transfer across cities without costly re-calibration. Explanation builds causal DAG frameworks that make AI decisions interpretable to transportation practitioners. We show the three vertices reinforce each other, and close with three golden rules for any deployable ITS model: handle long-tail cases safely, operate from day one, and explain itself to people with a transportation background. TrafficPDE is not a survey of what has been done. It is a guidebook for what must be built.

Explainable AI is now a mature and rapidly expanding field, with a wide range of methods for interpreting models and explaining their behaviour. Despite this progress, many of these methods have yet to make their way into the engineering pipelines where AI is actually deployed. This gap suggests a question that is central to this paper: what would it take for explanations to be informative, useful and dependable when deployed? I approach this question through counterfactual explanations, which I view as a promising foundation for practical XAI: by showing what would need to change for an AI decision to differ, they are easy to interpret and carry actionable information. In their standard form, however, counterfactuals come with limitations of form and scope that constrain their applicability in deployment. I therefore consider three lines of work that build on standard counterfactual explanations and ask what would need to change for them to be better aligned with deployment requirements. The first concerns robustness, ensuring that counterfactuals remain valid under the perturbations that deployment introduces. The second extends them beyond one-shot decisions to capture sequential decision-making. The third takes counterfactuals as a starting point for contestability, arguing that meaningful contestation requires more than an explanation alone. I close by reflecting on energy systems as a critical testbed for explainability, one that highlights the promises and limitations of current methods and may offer a source of criteria for the next generation of explanation methods.

We describe a research programme aimed at the construction of a single, generally intelligent agent - one competent across all tasks rather than skilled at any one. Such an agent must act in the real world through a rich sensorimotor interface, with sensors and actuators general enough for any task it may face. But that same richness makes any single task nearly impossible to learn directly. We argue that the resolution lies in abstraction, and survey our work towards this goal: discovering skills from experience, deriving compact models from those skills, and transferring both to new tasks. Skills, in particular, can be composed so that new tasks are solved with no further learning. Together, these threads work towards a single agent capable, in principle, of every task its body allows.

Recommender systems are evolving from models that predict user–item matching scores into intelligent systems that learn from relational structure, retrieve evidence, reason over user intent, and support personalized actions. This talk discusses a research agenda for trustworthy recommender systems in the era of agentic AI, organized around a transition from relational user modeling to generative personalization. We will discuss three connected components: relational user modeling with deep neural networks, generative and agentic personalization with large language models, and trustworthy foundation for reliable recommendations. The discussion revisits how social, behavioral, and knowledge relations can be integrated for user modeling; how generative models and personalized agents expand recommendation from ranking to intelligent decision support; and how trustworthiness challenges should be addressed throughout the system. The talk aims to provide a unified view of recommender systems as relational, generative, agentic, and trustworthy AI systems.

Artificial Intelligence (AI) and Machine Learning hold immense potential to accelerate scientific discovery and engineering design. A fundamental challenge in these domains involves efficiently exploring a large space of designs or hypotheses using expensive experiments in a resource-efficient manner. This paper surveys novel adaptive experimental design methods to address this broad challenge. Specifically, we discuss new probabilistic modeling and decision-making techniques that are applicable in small data settings. These approaches have shown substantial improvements in sample-efficiency, particularly for black-box optimization over high-dimensional combinatorial spaces (e.g., sequences and graphs) and a variety of goals ranging from multiobjective to multi-fidelity optimization. This paper outlines key methods and their real-world sustainability applications in areas such as nanoporous materials discovery, hardware design, surfactant design, and additive manufacturing.

Recent progress in Large Language Models (LLMs) has transformed text and code generation, yet models still falter on Partial Differential Equations (PDEs) where correctness, constraints, and physical consequences are critical. We explore how formal LLM reasoning can advance symbolic PDE modeling. First, our PDE-Controller formalizes informal PDEs, synthesizes solver-ready code, and plans subgoals to tackle nonconvex control via interactions with external solvers. Second, our Lean Finder accelerates PDE formalization via a semantics-aware search engine for Lean/Mathlib that retrieves relevant theorems, outperforming GPT models and gaining significant traction in the AI-for-math community. Through these efforts, we aim to design a semantics-first LLM that autoformalizes informal PDE problems into machine-checked specifications and synthesizes solver-ready code. This closes the loop between formal analysis and LLM reasoning, ultimately surpassing human heuristics across PDEs.

Large Language Models (LLMs) excel in natural language processing (NLP) tasks. However, they suffer from inherent limitations due to their sequence-based nature, such as structural information loss and factual unreliability. Graphs, with the ability to explicitly model entities and relations, offer an effective way to address these shortcomings. To systematically synthesize the emerging research on graph-enhanced LLMs, this survey, Graph4LLM, examines how these methods integrate graphs into various stages of the LLM pipeline, including the input, model, and output phases. For each phase, we provide a detailed review of the key methods and techniques. We also introduce a wide range of application scenarios where Graph4LLM methods demonstrate significant potential. Finally, we outline the challenges and future research directions for developing more efficient and interpretable solutions.

The advent of artificial intelligence has propelled AI-Generated Game Commentary (AI-GGC) into a rapidly expanding research area, offering advantages such as scalable availability and personalized narration. However, existing studies remain fragmented, and a systematic survey that unifies prior efforts is still lacking. To bridge this gap, our survey introduces a unified framework that systematically organizes the AI-GGC landscape. We present a novel taxonomy focused on three core commentator capabilities: Live Observation, Strategic Analysis, and Historical Recall, and further categorize commentary into three corresponding types: Descriptive Commentary, Analytical Commentary, and Background Commentary. Building on this structure, we provide an in-depth review of methods, datasets, and evaluation metrics, analyzing their strengths and limitations. Finally, we highlight key challenges and point out promising directions for future research in AI-GGC.