IJCAI 2026 Papers

Motivated by practical allocation problems, we study integer splittable congestion games with capacitated resources and player-specific costs. In this setting, each player has an integer weight that has to be split in integer units across multiple resources, each with a capacity limiting the total assigned weight, i.e., its congestion. The latency of a resource is equal to a player-specific constant when its congestion is not greater than the capacity and becomes prohibitive, i.e., equal to ∞, once the capacity is exceeded. We analyze the computational complexity of finding an allocation that optimizes utilitarian social welfare under two cost models (total cost and per-unit cost). Furthermore, we investigate the computation of, speed of convergence to, and efficiency of Nash equilibria.

IC3, also known as property-directed reachability (PDR), is a commonly-used algorithm for hardware safety model checking. It checks if a state transition system complies with a given safety property. IC3 either returns UNSAFE (indicating property violation) with a counterexample trace, or SAFE with a checkable inductive invariant as the proof to safety. In practice, the performance of IC3 is dominated by a large web of interacting heuristics and implementation choices, making manual tuning costly, brittle, and hard to reproduce. This paper presents IC3-Evolve, an automated offline code-evolution framework that utilizes an LLM to propose small, slot-restricted and auditable patches to an IC3 implementation. Crucially, every candidate patch is admitted only through proof-/witness-gated validation: SAFE runs must emit a certificate that is independently checked, and UNSAFE runs must emit a replayable counterexample trace, preventing unsound edits from being deployed. Since the LLM is used only offline, the deployed artifact is a standalone evolved checker with zero ML/LLM inference overhead and no runtime model dependency. We evolve on the public hardware model checking competition (HWMCC) benchmark and evaluate the generalizability on unseen public and industrial model checking benchmarks, showing that IC3-Evolve can reliably discover practical heuristic improvements under strict correctness gates.

History-dependent policies induced by recurrent neural networks (RNNs) rely on latent hidden state dynamics, making verification in partially observable reinforcement learning (RL) challenging. Existing RNN verification tools typically rely on restrictive modeling assumptions or coarse over-approximations of the hidden state space, which can lead to overly conservative or inconclusive results. We propose RNN Probabilistic Verification (RNN-ProVe), a probabilistic framework that estimates the likelihood of undesired behaviors in RNN-based policies. RNN-ProVe uses policy-driven sampling to approximate the set of hidden states that are feasible under a trained policy, and derives statistical error bounds to produce bounded-error, high-confidence estimates of behavioral violations. Experiments on partially observable single-agent and cooperative multi-agent tasks show that RNN-ProVe yields more quantitative, feasibility-aware probabilistic guarantees than existing tools, while scaling to recurrent and multi-agent settings.

Reinforcement learning (RL) agents often struggle to generalize to new tasks and contexts without updating their parameters, mainly because their learned representations and policies are overfit to the specifics of their training environments. To boost agents' in-context RL (ICRL) ability, this work formulates ICRL as a two-agent emergent communication problem and introduces CORAL (Communicative Representation for Adaptive RL), a framework that learns a transferable communicative context by functionally separating latent representation learning from control. In CORAL, an Information Agent (IA) is pre-trained as a world model on a diverse distribution of tasks. Its objective is not direct return maximization, but world modeling and distilling its understanding into concise messages. The emergent communication protocol is shaped by a novel Causal Influence Loss, which measures the effect that the message has on the next action. During deployment, the previously trained IA serves as a fixed contextualizer for a new Control Agent (CA), which learns to solve tasks by interpreting the provided communicative context. Our experiments demonstrate that this approach enables the CA to achieve significant gains in sample efficiency and successfully perform zero-shot adaptation with the help of pre-trained IA in diverse online and offline environments, validating the efficacy of learning a transferable communicative representation.

Understanding underlying agent attributes such as goals, preferences, beliefs, and ability level is key to explaining decision-making and predicting behaviour in complex environments. While goal recognition (GR) has produced effective methods for inferring goals, recent work has begun to explore reinforcement learning (RL)-based approaches to attribute recognition (AR), which generalises GR by inferring a wider range of agent attributes beyond goals. Existing frameworks treat attributes as discrete variables and typically require separate models for each attribute. In reality, many agent attributes vary continuously rather than as discrete values, making continuous representations crucial for capturing subtle variations in behaviour. We introduce a multivariate RL-based AR framework that jointly infers multiple continuous-valued attributes from observed behaviour using a single attribute-conditioned policy and continuous probabilistic inference. Our experiments demonstrate that the proposed framework achieves stable and fine-grained inference, offering improved flexibility and generality compared to existing approaches.

Graphical User Interface (GUI) Agents, powered by large language and vision-language models, hold promise for enabling end-to-end automation in digital environments. However, their progress is fundamentally constrained by the scarcity of scalable, high-quality trajectory data. Existing data collection strategies either rely on costly and inconsistent manual annotations or on synthetic generation methods that trade off between diversity and meaningful task coverage. To bridge this gap, we present GUI-ReWalk: a reasoning-enhanced, multi-stage framework for synthesizing realistic and diverse GUI trajectories. GUI-ReWalk begins with a stochastic exploration phase that emulates human trial-and-error behaviors, and progressively transitions into a reasoning-guided phase where inferred goals drive coherent and purposeful interactions. Moreover, it supports multi-stride task generation, enabling the construction of long-horizon workflows across multiple applications. By combining randomness for diversity with goal-aware reasoning for structure, GUI-ReWalk produces data that better reflects the intent-aware, adaptive nature of human-computer interaction. We further train Qwen2.5-VL-7B on the GUI-ReWalk dataset and evaluate it across multiple benchmarks, including Screenspot-Pro, OSWorld/OSWorld-G, UI-Vision, AndroidControl, and GUI-Odyssey. Results demonstrate that GUI-ReWalk enables superior coverage of diverse interaction flows, higher trajectory entropy, and more realistic user intent. These findings establish GUI-ReWalk as a scalable and data-efficient framework for advancing GUI agent research and enabling robust real-world automation.

Bound-Independent Pruning (BIP) is a powerful technique for accelerating tree-based complete search algorithms for Distributed Constraint Optimization Problems (DCOPs) by pruning dominated regions of the search space using only local information. However, BIP fundamentally relies on constructing a context-dependent dominating space through joint optimization over local constraints, incurring exponential computational and storage complexity that limits its practical scalability. To address this limitation, we propose DR-BIP, a decoupled and sound reformulation of BIP that enables efficient identification of dominated subspaces without explicitly computing the dominating space. By reformulating dominated-subspace detection into pairwise dominance tests and decoupling them across independent constraints, DR-BIP eliminates joint optimization and reduces both computational and storage complexity from exponential to polynomial. The resulting dominance components can be precomputed and reused across context changes, further improving efficiency. We formally prove the correctness of DR-BIP and integrate it into several state-of-the-art tree-based complete DCOP algorithms. Extensive empirical evaluation on standard benchmarks confirms that DR-BIP consistently outperforms BIP, significantly reducing both computation time and communication overhead.

In Partially Cooperative Markov Games (PCMG), agents need to learn effective cooperation under individual rewards, with the key challenge lies in leveraging these rewards to enhance team benefits optimally. Model aggregation (MA) is a promising solution due to its simplicity and efficiency, yet it suffers from instability caused by the volatility of individual policies trained with individual rewards and the non-asymptotic nature of aggregation updates. To address these challenges, we propose Two-Stage Model Aggregation (TSMA), a novel multi-policy MA framework for multi-agent reinforcement learning (MARL) that leverages individual policies to enhance team cooperation. Specifically, to enhance the stability and expressiveness of individual policies, each TSMA agent trains two individual policies (trained using the same individual reward) with parameter-level variation and combines them via adaptive weighted aggregation. The aggregated individual policy is then integrated with the team policy (trained using team reward), with integration weights dynamically adjusted over training. Additionally, L2 regularization is applied during the data-driven training to align their parameter distributions across policies, mitigating instability from non-asymptotic aggregation. As a general framework, TSMA is compatible with standard actor-critic MARL algorithms. We integrate TSMA with MAPPO and MADDPG, and evaluate them on challenging PettingZoo benchmarks, demonstrating significant improvements in PCMG. The code is available at: https://github.com/RL-DLMU/TSMA.

LLM-based role-playing has rapidly improved in fidelity, yet stronger adherence to persona constraints commonly increases vulnerability to jailbreak attacks, especially for risky or negative personas. Most prior work mitigates this issue with training-time solutions (e.g., data curation or alignment-oriented regularization). However, these approaches are costly to maintain as personas and attack strategies evolve, can degrade in-character behavior, and are typically infeasible for frontier closed-weight LLMs. We propose a training-free Dual-Cycle Adversarial Self-Evolution framework with two coupled cycles. A Persona-Targeted Attacker Cycle synthesizes progressively stronger jailbreak prompts, while a Role-Playing Defender Cycle distills observed failures into a hierarchical knowledge base of (i) global safety rules, (ii) persona-grounded constraints, and (iii) safe in-character exemplars. At inference time, the Defender retrieves and composes structured knowledge from this hierarchy to guide generation, producing responses that remain faithful to the target persona while satisfying safety constraints. Extensive experiments across multiple proprietary LLMs show consistent gains over strong baselines on both role fidelity and jailbreak resistance, and robust generalization to unseen personas and attack prompts.

Agents implemented using Large Language Models (LLMs) offer novel solutions for tasks such as document summarization. However, single-agent systems, limited by their own knowledge base, often exhibit information omissions or inconsistencies with the facts, while multi-agent systems lacking effective organization can fall into the trap of social cognitive defects such as flattery and premature consensus. Inspired by the U.S. jury system, we propose JURIS, a judicial-decision-inspired multi-agent collaborative framework. Specifically, the document summarization process simulates a lawsuit scenario: multiple generator agents condense the target text into multiple sentences from different perspectives, much like lawyers on both sides providing different evidence; multiple decision-making agents consider the overall picture based on defense and voting mechanisms to select candidate sentences, much like a jury adopting different evidence; finally, a chief editor agent compiles a summary of all selected sentences, much like a secretary summarizing a plan and a judge giving a final verdict. Extensive experiments on in-domain news and cross-domain benchmarks demonstrate that JURIS consistently and significantly outperforms strong single-agent and multi-agent baselines across automatic evaluation metrics, multi-dimensional quality assessments, and human judgments, validating the effectiveness of decision-driven structured multi-agent collaboration for document summarization.

Real-world multi-agent systems, from traffic coordination to resource allocation, are often modeled as general-sum games where individual incentives conflict with collective welfare. In these settings, the central challenge is not merely finding an equilibrium, but selecting socially desirable outcomes among many suboptimal Nash equilibria. Standard deep multi-agent reinforcement learning (MARL) methods struggle with this problem, as value-decomposition approaches are constrained by monotonicity assumptions and policy-gradient methods often converge to stable but socially inefficient equilibria. To address this limitation, we propose Phi-Actor-Critic (Phi-AC), a framework that leverages swap regret minimization to steer learning toward high-welfare correlated equilibria (CE). To make counterfactual regret estimation tractable in deep MARL, Phi-AC employs a centralized attention critic that predicts vector-valued regrets in a single forward pass, avoiding computationally expensive counterfactual simulations. We further introduce a Lagrangian-based equilibrium selection mechanism that optimizes social welfare while enforcing stability through regret constraints. Experiments on matrix games, Multi-Agent Particle Environments (MPE), and the Melting Pot Harvest scenario demonstrate that Phi-AC learns efficient and stable coordination strategies across diverse mixed-motive settings while maintaining high collective return and competitive fairness.

Existing Autonomous Mobility-on-Demand (AMoD) systems and demand-responsive algorithms are designed to adaptively respond to traffic demands, enhancing traffic service quality and profitability. However, such approaches may exacerbate traffic congestion by concentrating vehicles in high-demand areas. In addition, repeated braking, acceleration, and low-speed driving in traffic jams lead to substantial increases in emissions. Moreover, existing Deep Reinforcement Learning (DRL)-based approaches fail to address the instability of cooperative learning in environments with highly variable rewards, such as carbon emissions. To address these issues, we propose the Spatio-Temporal Attention Rebalancing (STAR) framework, leveraging an encoder-decoder architecture. In the proposed framework, we maximize the number of requests served and minimize CO₂ emissions, while ensuring high service quality by limiting maximum travel delays and waiting times for passengers. To evaluate the proposed framework, we conduct simulations in realistic urban traffic scenarios. Experimental results demonstrate that our framework consistently outperforms existing baselines. In a large-scale scenario with a fleet of one hundred vehicles, our approach improves the service rate by up to 10.1% and reduces CO₂ emissions per passenger by up to 69.6% compared to the baseline methods. The proposed framework and baselines are publicly available at https://github.com/2jungeuni/eco-friendly-fleet-rebalancing.

Many multiagent systems depend on collective decisions made by self-interested agents, which raises deep questions about coalition formation and stability. We study social choice with endogenous, outcome-contingent transfers, where agents voluntarily form contracts that redistribute utility depending on the collective decision, allowing for fully strategic coalition formation. We show that under consensus rules, individually rational strong Nash equilibria (IR–SNE) always exist, implementing welfare-maximizing outcomes with feasible transfers, and provide a simple, efficient algorithm to construct them. For more general anonymous, monotonic, and resolute rules, we identify necessary conditions for viable deviations, significantly limiting the possibility of destabilizing coalitions. By bridging cooperative and noncooperative perspectives, our approach shows that transferable utility can achieve core-like stability, restoring efficiency and budget balance even where classical impossibility results generally apply. Overall, this framework offers a practical and robust way to coordinate large-scale strategic multiagent systems.

We introduce and study the multi-agent stochastic shortest path (MSSP) problem, in which k agents strive to reach a target state, aiming to minimize the expected time to reach the target by any agent. We analyze the computational and strategy-complexity of the problem in both autonomous and coordinated settings, and we design efficient strategy-synthesis algorithms. The algorithms are experimentally evaluated on instances of increasing size against natural baselines.

We study EFX and approximate EFX allocations of indivisible goods among agents with at most k distinct valuations (i.e., k types of agents). For exact EFX, it is known that an EFX allocation always exists when there are at most three types of agents. For approximate EFX, the best known guarantee is a 0.618-EFX allocation, and this has been improved recently to a φ-EFX allocation when the number of agents is at most seven. We settle another natural case in this landscape by showing that a φ-EFX allocation exists for any number of agents whenever there are at most four distinct valuations. We then consider a relaxation, EFX with charity, where some goods may remain unallocated and no agent envies the set of unallocated goods. It is known that, for n agents and any ε ∈ (0, 1/2], there exists an EFXε allocation with at most Õ((n/ε)^(1/2)) goods allocated to charity. We prove that when there are at most k distinct valuation types, there exists an EFXε allocation with only Õ((k/ε)^(1/2)) goods allocated to charity; in particular, the required amount of charity is sublinear in the number of distinct valuations rather than in the number of agents. We show that any EFX guarantee for k types of agents can be extended, with an additional (1 − 2ε) approximation factor, to a more general setting where the valuations can be partitioned into k clusters such that the maximum heterogeneity of valuations within each cluster is at most ε.

Clinical Computed Tomography (CT) reconstruction often faces mixed degradations, where quantum noise, streak artifacts, and geometric distortions co-occur with various compositions and severities. Recently, all-in-one frameworks have outperformed traditional single-task models through degradation-specialized modulation mechanisms. However, attempting to resolve these mixed degradations simultaneously forces the model into suboptimal trade-offs, failing to balance conflicting reconstruction objectives. Our empirical study shows that decomposing CT reconstruction into a sequence of ordered steps effectively mitigates this conflict, with the execution order being a critical performance factor. Leveraging this insight, we propose AgenticCT, an LLM-orchestrated multi-agent framework designed to autonomously plan the optimal reconstruction trajectory. Specifically, AgenticCT operates through a diagnose-plan-treat workflow: a Supervisor Agent explicitly estimates a structured degradation state; a Planner Agent then selects a topology-optimized execution order conditioned on this state; and a library of Processor Agents sequentially executes the trajectory to achieve high-fidelity reconstruction. Extensive experiments demonstrate that AgenticCT consistently improves reconstruction fidelity across single and mixed degradations, and generalizes better to out-of-distribution datasets compared with strong specialized and all-in-one baselines. Code is available at: https://github.com/yqhuang2912/AgenticCT.

Tool learning methods have enhanced the ability of large language models (LLMs) to interact with real-world applications. Many existing works fine-tune LLMs or design prompts to enable LLMs to select appropriate tools and correctly invoke them to meet user requirements. However, it is observed in previous works that the performance of tool learning varies from tasks, datasets, training settings, and algorithms. Without understanding the impact of these factors, it can lead to inconsistent results, inefficient model deployment, and suboptimal tool utilization, ultimately hindering the practical integration and scalability of LLMs in real-world scenarios. Therefore, in this paper, we explore the impact of both internal and external factors on the performance of tool learning frameworks. Through extensive experiments on two benchmark datasets, we find several insightful conclusions for future work, including the observation that LLMs can benefit significantly from increased trial and exploration. We believe our empirical study provides a new perspective for future tool learning research.

Empowered by large language models (LLMs), multi-agent systems (MAS) have shown significant potential in code generation by simulating collaborative workflows. However, we identify a collaboration degeneration phenomenon, where one agent dominates while others remain disengaged, occurring in 38.4% of non-routine HumanEval tasks. Surprisingly, introducing competition among LLM agents eliminates this degeneration and sparks innovation in generated code. Inspired by this catfish effect, we posit that competition helps prevent collaboration degeneration. We thus propose C3, a Controllable Competitive Collaboration framework featuring two novel mechanisms: Centralized Auction-Based Competition (CAB) for role allocation via bidding and Decentralized Communication-Aware Competition (DCC) for solution refinement through opponent-aware adaptation. Evaluated on 70 complex projects from extended SoftwareDev, C3 reduces collaboration degeneration from 32.4% to 12.8%, while enhancing code innovation and diversity. Human evaluations further confirm C3’s superiority in functionality, maintainability, and robustness over purely collaborative or competitive baselines.

Transformer models are rapidly becoming a cornerstone of modern Internet of Things (IoT) applications, yet their computational and memory demands far exceed the capabilities of a single typical ultra-low-power IoT device. We present CATS, a framework for distributed transformer inference on ultra-low-power wireless devices, enabling multiple devices to collaboratively execute models far larger than what a single device can sustain. At its core, CATS is a communication-aware distributed transformer inference scheme co-designed across transformer partitioning, wireless communication and training. It employs SomeGather, a new pruned communication primitive that selectively broadcasts activation columns to reduce communication bandwidth and RAM usage without sacrificing model accuracy. Building on SomeGather, we design a partitioning method that exploits this primitive for efficient model parallelism. To cope with unreliable wireless communication, CATS employs message-dropout during training, which mimics packet losses and yields models that are robust to message loss during inference. In real-world experiments, we show that CATS brings distributed transformer inference to ultra-low-power wireless devices for the first time, with deployments on up to 16 devices that collaboratively execute transformer models up to 14 times larger than what a single device can run.