ICLR 2026 论文列表

The explosive growth of large language models (LLMs) has created a vast but opaque landscape: millions of models exist, yet their evolutionary relationships through fine-tuning, distillation, or adaptation are often undocumented or unclear, complicating LLM management. Existing methods are limited by task specificity, fixed model sets, or strict assumptions about tokenizers or architectures. Inspired by biological DNA, we address these limitations by mathematically defining *LLM DNA* as a low-dimensional, bi-Lipschitz representation of functional behavior. We prove that LLM DNA satisfies *inheritance* and *genetic determinism* and establish its existence. Building on this theory, we derive a general, scalable, training-free pipeline for DNA extraction. In experiments across 305 LLMs, DNA aligns with prior studies on limited subsets and achieves superior or competitive performance on various tasks. Beyond these tasks, DNA comparisons uncover previously undocumented relationships among LLMs. We further construct the evolutionary tree of LLMs using phylogenetic algorithms, which align with shifts from encoder-decoder to decoder-only architectures, reflect temporal progression, and reveal distinct evolutionary speeds across LLM families.

Current automated agent design frameworks produce either static workflows that lack adaptability or per-query optimizers that prevent the accumulation of deep, agent-level task expertise. We propose a new direction that reconciles these paradigms: creating stateful teams of specialist agents that accumulate knowledge over time and can be reconfigured for novel tasks entirely without human intervention. To this end, we introduce \textsc{ASpec}, a framework that manages this full agent lifecycle by first autonomously $\textbf{discovering}$ specialist archetypes via evolutionary search and then $\textbf{cultivating}$ their expertise through experience, mirroring how human experts learn through practice and reflection. We further introduce a lightweight hierarchical control policy, "retain-then-escalate," which governs when to leverage the established agent system versus when to adapt its structure. Through comprehensive experiments, we demonstrate that this approach leads to significant performance gains on expert-level scientific benchmarks like GPQA while matching the state-of-the-art on broader domain tasks, demonstrating a promising path toward agent systems that are simultaneously expert, adaptive, and efficient.

Tool agents interact with users through multi-turn dialogues to accomplish various tasks. Recent studies have adopted user simulation methods to develop these agents in multi-turn settings. However, existing user simulators tend to be agent-friendly, exhibiting only cooperative behaviors, which fails to train and test agents against non-collaborative users in the real world. To address this, we propose a novel user simulator architecture that simulates four categories of non-collaborative behaviors: requesting unavailable services, digressing into tangential conversations, expressing impatience, and providing incomplete utterances. Our user simulator can simulate challenging and natural non-collaborative behaviors while reliably delivering all intents and information necessary to accomplish the task. Our experiments on MultiWOZ and $\tau$-bench reveal significant performance degradation in state-of-the-art tool agents when encountering non-collaborative users. We provide detailed analyses of agents' weaknesses under each non-collaborative condition, such as escalated hallucinations and dialogue breakdowns. Ultimately, we contribute an easily extensible user simulation framework to help the research community develop tool agents and preemptively diagnose them under challenging real-world conditions within their own services.

Multi-agent reinforcement learning (MARL) has made significant progress in recent years, but most algorithms still rely on a discrete-time Markov Decision Process (MDP) with fixed decision intervals. This formulation is often ill-suited for complex multi-agent dynamics, particularly in high-frequency or irregular time-interval settings, leading to degraded performance and motivating the development of continuous-time MARL (CT-MARL). Existing CT-MARL methods are mainly built on Hamilton–Jacobi–Bellman (HJB) equations. However, they rarely account for safety constraints such as collision penalties, since these introduce discontinuities that make HJB-based learning difficult. To address this challenge, we propose a continuous-time constrained MDP (CT-CMDP) formulation and a novel MARL framework that transforms discrete MDPs into CT-CMDPs via an epigraph-based reformulation. We then solve this by proposing a novel PINN-based actor–critic method that enables stable and efficient optimization in continuous time. We evaluate our approach on continuous-time safe multi-particle environments (MPE) and safe multi-agent MuJoCo benchmarks. Results demonstrate smoother value approximations, more stable training, and improved performance over safe MARL baselines, validating the effectiveness and robustness of our method.

Powerful large language models (LLMs) from different providers have been expensively trained and finetuned to specialize across varying domains. In this work, we introduce a new kind of Conductor model trained with reinforcement learning to automatically discover powerful coordination strategies among LLMs. Our Conductor learns not only to design targeted communication topologies for effective agent-to-agent collaboration, but also to prompt engineer focused instructions to the LLMs to maximally leverage their individual capabilities. We show that, by learning optimal coordination strategies over pools of powerful worker LLMs, a 7B Conductor achieves significant performance gains beyond any individual worker, attaining state-of-the-art results in challenging reasoning benchmarks, such as LiveCodeBench and GPQA. By training with randomized agent pools, our conductor effectively adapts to arbitrary sets of open- and closed-source agents, meeting any user requirements. Furthermore, allowing the Conductor to select itself as a worker gives rise to recursive topologies, elevating performance with a new form of dynamic test-time scaling through online iterative adaptation. More broadly, ours is among the early work demonstrating language model coordination can be unlocked through RL, where powerful coordination strategies emerge naturally in LLMs through pure end-to-end reward maximization.

High-quality time series (TS) data are essential for ensuring TS model performance, rendering research on rating TS data quality indispensable. Existing methods have shown promising rating accuracy within individual domains, primarily by extending data quality rating techniques such as influence functions and Shapley values to account for temporal characteristics. However, they neglect the fact that real-world TS data can span vastly different domains and exhibit distinct properties, hampering the accurate and efficient rating of diverse TS data. In this paper, we propose TSRating, a novel and unified framework for rating the quality of time series data crawled from diverse domains. TSRating leverages LLMs' inherent ample knowledge, acquired during their extensive pretraining, to comprehend and discern quality differences in diverse TS data. We verify this by devising a series of prompts to elicit quality comparisons from LLMs for pairs of TS samples. We then fit a dedicated rating model, termed TSRater, to convert the LLMs' judgments into efficient quality predictions by inferring future TS samples through TSRater's inference. To ensure cross-domain adaptability, we develop a meta-learning scheme to train TSRater on quality comparisons collected from nine distinct domains. To improve training efficiency, we employ signSGD for inner-loop updates, thus circumventing the demanding computation of hypergradients. Extensive experimental results on eleven benchmark datasets across three time series tasks, each using both conventional TS models and TS foundation models, demonstrate that TSRating outperforms baselines in terms of estimation accuracy, efficiency, and domain adaptability.

In cooperative multi-agent reinforcement learning, agents often face scalability challenges due to the exponential growth of the joint action and observation spaces. Inspired by the structure of human teams, we explore subteam-based coordination, where agents are partitioned into fully correlated subgroups with limited inter-group interaction. We formalize this structure using Bayesian networks and propose a class of correlated joint policies induced by directed acyclic graphs . Theoretically, we prove that regularized policy gradient ascent converges to near-optimal policies under a decomposability condition of the environment. Empirically, we introduce a heuristic for dynamically constructing context-aware subteams with limited dependency budgets, and demonstrate that our method outperforms standard baselines across multiple benchmark environments.

Value decomposition is a core approach for cooperative multi-agent reinforcement learning (MARL). However, existing methods still rely on a single optimal action and struggle to adapt when the underlying value function shifts during training, often converging to suboptimal policies. To address this limitation, we propose Successive Sub-value Q-learning (S2Q), which learns multiple sub-value functions to retain alternative high-value actions. Incorporating these sub-value functions into a Softmax-based behavior policy, S2Q encourages persistent exploration and enables $Q^{\text{tot}}$ to adjust quickly to the changing optima. Experiments on challenging MARL benchmarks confirm that S2Q consistently outperforms various MARL algorithms, demonstrating improved adaptability and overall performance. Our code is available at https://github.com/hyeon1996/S2Q.

Large language model (LLM)-based agents increasingly rely on tool use to complete real-world tasks. While existing works evaluate the LLMs' tool use capability, they largely focus on the final answers yet overlook the detailed tool usage trajectory, i.e., whether tools are selected, parameterized, and ordered correctly. We introduce TRAJECT-Bench, a trajectory-aware benchmark to comprehensively evaluate LLMs' tool use capability through diverse tasks with fine-grained evaluation metrics. TRAJECT-Bench pairs high-fidelity, executable tools across practical domains with tasks grounded in production-style APIs, and synthesizes trajectories that vary in breadth (parallel calls) and depth (interdependent chains). Besides final accuracy, TRAJECT-Bench also reports trajectory-level diagnostics, including tool selection and argument correctness, and dependency/order satisfaction. Analyses reveal failure modes such as similar tool confusion and parameter-blind selection, and scaling behavior with tool diversity and trajectory length where the bottleneck of transiting from short to mid-length trajectories is revealed, offering actionable guidance for LLMs' tool use.

Large-scale reinforcement learning with verifiable rewards (RLVR) has proven effective in harnessing the potential of large language models (LLMs) for single-turn reasoning tasks. In realistic reasoning scenarios, LLMs often rely on external tools to assist in task-solving processes. However, current RL algorithms typically employ trajectory-level rollout sampling, consistently neglecting the fine-grained exploration of multi-turn tool-call steps. To bridge this gap, we propose Agentic Reinforced Policy Optimization (ARPO), a novel agentic RL algorithm tailored for training multi-turn LLM-based agents. Our preliminary experiments reveal that LLMs frequently exhibit increased uncertainty after tool-call steps, as evidenced by higher entropy in the distribution of generated tokens. Motivated by this, ARPO incorporates an entropy-based adaptive rollout mechanism, encouraging the policy model to adaptively branch sampling during high-entropy tool-call rounds, thereby promoting step-level exploration of latent tool-use behaviors. By integrating an advantage attribution estimation, ARPO enables LLMs to internalize advantage differences in stepwise tool-use interactions. Experiments across 13 challenging benchmarks demonstrate ARPO's superiority over trajectory-level RL algorithms. Remarkably, ARPO achieves improved performance using only half of the tool-use budget required by existing methods, offering a scalable solution for aligning LLM-based agents with real-time dynamic environments. Our codes are released at https://github.com/RUC-NLPIR/ARPO.

Discovering scaling laws for predicting model performance at scale is a fundamental and open-ended challenge, mostly reliant on slow, case specific human experimentation. To investigate the potential for LLMs to automate this process, we collect over 5,000 experiments from existing literature and curate seven diverse scaling law discovery tasks. While existing agents struggle to produce accurate law formulas, this paper introduces SLDAgent, an evolution-based agent that co-optimize the scaling law model and the parameters, enabling it to autonomously explore complex relationships between variables. For the first time, we demonstrates that SLDAgent can automatically discover laws that exhibit consistently more accurate extrapolation than their established, human-derived counterparts across all tasks. Through comprehensive analysis, we elucidate why these discovered laws are superior and verify their practical utility in both pretraining and finetuning applications. This work establishes a new paradigm for agentic scientific discovery, showing that AI systems can understand their own scaling behavior, and can contribute novel and practical knowledge back to the research community.

Group-based reinforcement learning (RL), such as GRPO, has advanced the capabilities of large language models on long-horizon agentic tasks. To enable more fine-grained policy updates, recent research has increasingly shifted toward stepwise group-based policy optimization, which treats each step in a rollout trajectory independently while using a memory module to retain historical context. However, we find a key issue in estimating stepwise relative advantages, namely context inconsistency, where steps within the same group may differ in their historical contexts. Empirically, we reveal that this issue can lead to severely biased advantage estimation, thereby degrading policy optimization significantly. To address the issue, in this paper, we propose Hierarchical-of-Groups Policy Optimization (HGPO) for long-horizon agentic tasks. Specifically, within a group of rollout trajectories, HGPO assigns each step to multiple hierarchical groups according to the consistency of historic contexts. Then, for each step, HGPO computes distinct advantages within each group and aggregates them with an adaptive weighting scheme. In this way, HGPO can achieve a favorable bias-variance trade-off in stepwise advantage estimation, without extra models or rollouts. Evaluations on two challenging agentic tasks, ALFWorld and WebShop with Qwen2.5-1.5B-Instruct and Qwen2.5-7B-Instruct, show that HGPO significantly outperforms existing agentic RL methods under the same computational constraints.

Recent studies operationalize self-improvement through coding agents that edit their own codebases, grow a tree of self-modifications through expansion strategies that favor higher software engineering benchmark performance, considering that this implies more promising subsequent self-modifications. However, we identify a mismatch between the agent’s self-improvement potential (metaproductivity) and its coding benchmark performance, namely the \emph{Metaproductivity-Performance~Mismatch}. Inspired by Huxley’s concept of clade, we propose a metric ($\mathrm{CMP}$) that aggregates the benchmark performances of the \emph{descendants} of an agent as an indicator of its potential for self-improvement. We show that the G\"odel Machine, the optimal self-improving machine, is achieved with access to true $\mathrm{CMP}$. We introduce the Huxley-G\"odel Machine (HGM), which, by estimating $\mathrm{CMP}$ and using it as guidance, searches the tree of self-modifications. On SWE-bench Verified and Polyglot, HGM outperforms prior self-improving coding agent search methods while using less wall-clock time. Moreover, the agent optimized by HGM on SWE-bench Verified outperforms SWE-agent, a leading human-engineered open source coding agent on SWE-bench Lite, where SWE-agent ranks the best on the official leaderboard, when both use the GPT-5-mini backbone, demonstrating that HGM self-improvement indeed enhances genuine coding capability.

Generating interactive 3D scenes from text requires not only synthesizing assets but arranging them with spatial intelligence—support, affordances, and plausibility. However, training data for interactive scenes is dominated by a few indoor datasets, so learning-based methods overfit to in-distribution layouts and struggle to compose diverse arrangements (e.g., outdoor settings and small-on-large relations). Meanwhile, LLM-based layout planners can propose diverse arrangements, but the lack of visual grounding often yields implausible placements that violate commonsense physics. We propose Scenethesis, a training-free, agentic framework that couples LLM-based scene planning with vision-guided layout refinement. Given a text prompt, Scenethesis first drafts a coarse layout with an LLM; a vision module refines the layout and extracts scene structure to capture inter-object relations. A novel optimization stage enforces pose alignment and physical plausibility, and a final judge verifies spatial coherence and triggers targeted repair when needed. Across indoor and outdoor prompts, Scenethesis produces realistic, relation-rich, and physically plausible 3D interactive scenes, reducing collisions and stability failures compared to SOTA methods, making it practical for virtual content creation, simulation, and embodied AI.

The tendency to find and exploit "shortcuts" to complete tasks poses significant risks for reliable assessment and deployment of large language models (LLMs). For example, an LLM agent with access to unit tests may delete failing tests rather than fix the underlying bug. Such behavior undermines both the validity of benchmark results and the reliability of real-world LLM coding assistant deployments. To quantify, study, and mitigate such behavior, we introduce ImpossibleBench, a benchmark framework that systematically measures LLM agents' propensity to exploit test cases. ImpossibleBench creates "impossible" variants of tasks from existing benchmarks like LiveCodeBench and SWE-bench by introducing direct conflicts between the natural-language specification and the unit tests. We measure an agent's "cheating rate" as its pass rate on these impossible tasks, where any pass necessarily implies a specification-violating shortcut. As a practical framework, ImpossibleBench is not just an evaluation but a versatile tool. We demonstrate its utility for: (1) studying model behaviors, revealing more fine-grained details of cheating behaviors from simple test modification to complex operator overloading; (2) context engineering, showing how prompt, test access and feedback loop affect cheating rates; and (3) developing monitoring tools, providing a testbed with verified deceptive solutions. We hope ImpossibleBench serves as a useful framework for building more robust and reliable LLM systems.

This paper studies how to achieve accurate modeling and effective control in stochastic nonlinear dynamics with multiple interacting objects. However, non-uniform interactions and random topologies make this task challenging. We address these challenges by proposing Graph Controllable Embeddings (GCE), a general framework to learn stochastic multi-object dynamics for linear control. Specifically, GCE is built on Hilbert space embeddings, allowing direct embedding of probability distributions of controlled stochastic dynamics into a reproducing kernel Hilbert space (RKHS), which enables linear operations in its RKHS while retaining nonlinear expressiveness. We provide theoretical guarantees on the existence, convergence, and applicability of GCE. Notably, a mean field approximation technique is adopted to efficiently capture inter-object dependencies and achieves provably low sample complexity. By integrating graph neural networks, we construct data-dependent kernel features which are capable of adapting to dynamic interaction patterns and generalizing to even unseen topologies with only limited training instances. GCE scales seamlessly to multi-object systems of varying sizes and topologies. Leveraging the linearity of Hilbert spaces, GCE also supports simple yet effective control algorithms for synthesizing optimal sequences. Experiments on physical systems, robotics, and power grids validate GCE and demonstrate consistent performance improvement over various competitive embedding methods in both in-distribution and few-shot tests.

When are multi-agent LLM systems merely a collection of individual agents versus an integrated collective with higher-order structure? We introduce an information-theoretic framework to test---in a purely data-driven way---whether multi-agent systems show signs of higher-order structure. This information decomposition lets us measure whether dynamical emergence is present in multi-agent LLM systems, localize it, and distinguish spurious temporal coupling from performance-relevant cross-agent synergy. We implement both a practical criterion and an emergence capacity criterion operationalized as partial information decomposition of time-delayed mutual information (TDMI). We apply our framework to experiments using a simple guessing game without direct agent communication and only minimal group-level feedback with three randomized interventions. Groups in the control condition exhibit strong temporal synergy but only little coordinated alignment across agents. Assigning a persona to each agent introduces stable identity-linked differentiation. Combining personas with an instruction to ``think about what other agents might do'' shows identity-linked differentiation and goal-directed complementarity across agents. Taken together, our framework establishes that multi-agent LLM systems can be steered with prompt design from mere aggregates to higher-order collectives. Our results are robust across emergence measures and entropy estimators, and not explained by coordination-free baselines or temporal dynamics alone. Without attributing human-like cognition to the agents, the patterns of interaction we observe mirror well-established principles of collective intelligence in human groups: effective performance requires both alignment on shared objectives and complementary contributions across members.

Formal models are essential to specifying large, complex computer systems and verifying their correctness, but are notoriously expensive to write and maintain. Recent advances in generative AI show promise in generating certain forms of specifications. However, existing work mostly targets small code, not complete systems. It is unclear whether AI can deal with realistic system artifacts, as this requires abstracting their complex behavioral properties into formal models. We present SysMoBench, a benchmark that evaluates AI's ability to formally model large, complex systems. We focus on concurrent and distributed systems, which are keystones of today's critical computing infrastructures, encompassing operating systems and cloud infrastructure. We use TLA+, the de facto specification language for concurrent and distributed systems, though the benchmark can be extended to other specification languages. We address the primary challenge of evaluating AI-generated models by automating metrics like syntactic and runtime correctness, conformance to system code, and invariant correctness. SysMoBench currently includes eleven diverse system artifacts: the Raft implementation of Etcd and Redis, the leader election of ZooKeeper, the Spinlock, Mutex and Ringbuffer in Asterinas OS, etc., with more being added. SysMoBench enables us to understand the capabilities and limitations of today's LLMs and agents, putting tools in this area on a firm footing and opening up promising new research directions.

The chemical reaction recommendation is to select proper reaction condition parameters for chemical reactions, which is pivotal to accelerating chemical science.With the rapid development of large language models (LLMs), there is growing interest in leveraging their reasoning and planning capabilities for reaction condition recommendation.Despite their success, existing methods rarely explain the rationale behind the recommended reaction conditions, limiting their utility in high-stakes scientific workflows. In this work, we propose ChemMAS, a multi-agent system that reframes condition prediction as an evidence-based reasoning task. ChemMAS decomposes the task into mechanistic grounding, multi-channel recall, constraint-aware agentic debate, and rationale aggregation. Each decision is backed by interpretable justifications grounded in chemical knowledge and retrieved precedents. Experiments show that ChemMAS achieves 20–35\% gains over domain-specific baselines and outperforms general-purpose LLMs by 10–15\% in Top-1 similarity, while offering falsifiable, human-trustable rationales, which establishes a new paradigm for explainable AI in scientific discovery.