NeurIPS 2026 Papers

For groups of autonomous agents to achieve a particular goal, they must engage in coordination and long-horizon reasoning. Rather than relying on complex reward functions and explicit cooperation mechanisms, we ask what minimal ingredients are required for effective coordination and exploration to emerge in multi-agent settings. We investigate this question through self-supervised goal-reaching, where agents aim to maximize the likelihood of visiting a goal state rather than maximizing a reward. Despite a sparse feedback signal, we present empirical results that show self-supervised goal-reaching techniques enable agents to learn from such feedback. On MARL benchmarks, self-supervised goal-reaching outperforms alternative approaches that have access to the same sparse reward signal. Furthermore, we empirically demonstrate that multi-agent self-supervised goal-reaching approaches can be more robust than single-agent strategies. While there is no *explicit* exploration mechanism, this approach explores nontrivial intermediate coordination strategies in sparse settings where alternative approaches fail to achieve a single success.

Group-relative RL training (GRPO) samples a small group of parallel rollouts for every training prompt and uses their within-group reward spread to compute per-trajectory advantages. In agentic environments each rollout is a long multi-turn dialogue with one LLM call per step, so this multi-sample multiplier dominates the total training cost. When every rollout of a prompt ends with the same reward, the group has zero reward variance and contributes no gradient, so the extra rollouts add no information; such groups are common in practice (typically around 40% of all groups), so the wasted-compute fraction is substantial rather than marginal. Existing methods filter such groups at the prompt level, either after their rollouts are paid for or before any rollout begins, but both decide without using information that becomes available during the rollout itself. We instead ask whether the in-group divergence between the partial trajectories at an intermediate step can already predict that the group will be zero-variance: when the parallel rollouts have already converged on the same action prefix, the group is on track to produce a single reward, and we can stop early. We propose a one-parameter gate that stops a group when the mean pairwise prefix edit distance between its partial action sequences falls below a threshold. On a 60-iteration on-policy GRPO run on ALFWorld with Qwen2.5-7B, averaged over four random seeds, the gated arm finishes 10.7% faster in wall-clock (bootstrap 95% CI excludes 0) and shifts held-out success rate on 50 unseen tasks by +2.5 pp, with the held-out gain tracing to a measurable reduction in zero-advantage gradient-batch dilution.

Selecting a next motion in a populated scene is not just a matter of avoiding what is currently in the way: an action can be locally feasible yet drift onto the road instead of the crosswalk, cut into a stream of crossing pedestrians, or bottleneck a narrow passage. These failure modes are invisible from the current state alone, and acting on them requires reasoning about what each candidate action will cause. World models offer exactly this, namely action-conditioned predictions of the future, but rolling them out at every decision step is a cost deployable navigation systems cannot pay, and the field has largely traded foresight for speed. We argue this is a false dichotomy. The natural place for a world model in human-aware navigation is at training time, as a preference oracle that distills foresight into a fast policy, rather than at inference time, as an online planner. We pair a discrete visual action prior with two task-aligned world models, one for nearby social dynamics and one for future scene semantics, that score candidate actions by their predicted consequences and refine the policy through group-relative preference distillation. At deployment, both world models are removed and the policy acts in a single forward pass. Across five held-out scenes, this yields navigation that is simultaneously safer, more socially and group-coherent, and more goal-directed than strong reactive and learned baselines, with naive observers rating the resulting motion as more human-like. Taken together, our findings point to a different approach for bringing world-model foresight into deployable agents: not accelerating rollouts, but amortizing them into the policy so that the cost of looking ahead is confined to training.

The rapid emergence of generative image models has led to the development of specialized watermarking techniques, particularly in-generation methods such as seed-based embedding. However, current evaluations in this area remain largely empirical, making them heavily reliant on the specific model architectures used for generation and inversion. This prevents any clear conclusion on the performance of any method, especially regarding security, for which a rigorous definition is lacking. Against this approach, we argue that the effectiveness of a watermarking scheme should be established purely through a thorough theoretical analysis. This is enabled by decoupling the model-dependent part from the actual decision mechanism of the watermarking system. Using this decoupling, we introduce a formal evaluation framework based on security, robustness, and fidelity. This allows precise comparisons between watermarking systems through a characteristic surface representing the trade-off between these three quantities, independent of any generative model. Based on this framework, we propose SSB, a novel watermarking method that generalizes previous seed-based methods by allowing to reach any security-robustness-fidelity regime on its characteristic surface. This work opens the door to the design of modern watermarking systems with theoretical guarantees that do not necessitate any costly empirical evaluations.

By combining cloud-side reasoning with edge-side observation and execution, cloud-edge collaboration has emerged as a promising paradigm for mobile UI agents. However, existing approaches face a fundamental trade-off between cost and accuracy. On one hand, step-by-step cloud interaction ensures high accuracy by uploading UI screens at every step and leveraging the cloud's powerful reasoning capabilities, but this inevitably incurs significant privacy costs for edge data and high computational expenses on the cloud. On the other hand, one-shot cloud planning drastically reduces the demand for edge privacy data and cloud compute by generating a complete plan based solely on the initial UI screen. Yet, this often leads to a substantial drop in accuracy, as the cloud agent may rely on invalid environment assumptions such as unseen app capabilities or UI flows. In this paper, we identify that the root cause of this trade-off lies in the environment assumption gap: the cloud agent lacks prior knowledge of the environment information required for edge-side task execution, and is thus forced to choose between planning with real-time information and planning based on speculative assumptions. To tackle this dilemma, we propose SeaPilot, which proactively provides the cloud agent with the necessary environmental context at the initial task submission stage, thereby achieving the dual benefits of both step-by-step and one-shot methods. To realize this, SeaPilot employs an iterative self-refining mechanism that progressively acquires feedback knowledge from execution failures across diverse tasks, enabling it to learn how to supply precise environmental information for each task in advance. Empirically, SeaPilot improves accuracy by 46.7% over one-shot methods, reduces cloud-token usage by 23.6x compared with step-by-step methods, and reduces privacy cost by 89.6%, achieving a better cost--accuracy trade-off. The code is available at https://anonymous.4open.science/r/SeaPilot-C13C.

Sparse rewards make cooperative multi-agent reinforcement learning difficult because useful feedback often appears only after multiple agents coordinate over long horizons. We introduce Multi-Agent Contrastive Reinforcement Learning (MACRL), a goal-conditioned method that trains parameter-shared decentralized actors with a centralized contrastive critic. Instead of learning from sparse task rewards, the critic contrasts joint state-action embeddings with future achieved goals from replay, turning every trajectory into dense self-supervised signal. Across tasks that vary in horizon length, coordination difficulty, and object interaction, our method is competitive with strong HER-based baselines when exploration is sufficient and substantially more robust when sparse rewards become uninformative. On the hardest tasks, it obtains substantial non-zero success while non-contrastive baselines remain near zero. Scaling from 4 to 64 layers further improves performance on several hard tasks, with gains up to 25 percentage points, while providing little consistent benefit to MA-SAC+HER. These results suggest that contrastive objectives and depth-scaled residual networks provide a promising foundation for scalable cooperative goal-conditioned control.

Multi-view 3D reconstruction is increasingly driven by feed-forward models, which are fast and robust but often imprecise or globally inconsistent across views, especially in sparse-view and low-overlap settings. Structure-from-motion (SfM) offers a complementary geometric signal through camera poses and sparse 3D points estimated by correspondence filtering and optimization. To propagate this globally consistent \emph{SfM scaffold} into dense geometry, we introduce \textbf{Scaffold3D}, an SfM-conditioned reconstruction framework that injects image patch-aligned SfM point tokens into a pairwise feed-forward pointmap predictor. Our approach preserves image-token reasoning while using explicit SfM structure to condition dense prediction. Since pairwise pointmaps are predicted in local frames, we fuse them with global alignment and an SfM anchoring term. Across established benchmarks (ScanNet++, ETH3D, and Tanks\&Temples) and out-of-domain data (4D-DRESS and MV-dVRK), Scaffold3D achieves stronger overall performance than feed-forward reconstruction models, point-token propagation, depth-completion baselines, and recent geometry-conditioned models. The gains are especially clear in low- and no-overlap evaluations. Our two-view SfM-conditioned pointmap predictor also outperforms several multi-view geometry-conditioned baselines, indicating that how geometric inputs are fused with image-token reasoning is as important as the number of conditioned views. Together, these results establish SfM-scaffold conditioning as a practical interface between learned pointmap prediction and classical geometric optimization.

A vision-language (VL) agent wraps a frozen vision-language model (VLM) into a test-time system that may call external tools during execution before producing a response. The central problem is how the VLM and its tools interact across multiple steps. Two recipes are common, both rigid. Thinking for longer (borrowed from text-only agents) sharpens reasoning, but without in-loop visual refresh, it lets visual evidence decay into hallucination. Querying tools more often, as traditional agents do, treats outputs as ground truth, a rigid commitment that fails when calls are noisy. We propose Saliency-Aware Principle Selection (SAP), a training-free inference-time method that avoids both rigidities: it searches a broad space of high-level principles, short textual directives (e.g., ``re-examine the image whenever an intermediate conclusion is formed'') that each prescribe a different way for the VLM to weigh tool advice across steps, and selects the principle whose parallel reasoning routes best fit the actual visual evidence. Tool outputs are treated as advisory (consulted, never authoritative), and a population-based evolutionary loop drives the search. Under the same token budget as long chain-of-thought, SAP substantially reduces object hallucination while remaining competitive on reasoning-heavy benchmarks. Being training-free and composed of independent routes, SAP is plug-and-play on any VLM and parallel for modern agent deployment.

Safe multi-agent reinforcement learning usually constrains team-level safety cost, but a safe team can still assign the same agent to repeatedly wait, yield, detour, absorb intervention, or lose access to a scarce resource. We study this failure mode as *burden unfairness under hard safety*. Safe-Fair multi-agent constrained policy optimization (Safe-Fair MACPO) augments MACPO with agent-wise burden logging, a temporal fairness-debt state, a fairness critic, and a two-cost trust-region update. Safety is lexicographically primary: recovery may override fairness to avoid unsafe actions, while the resulting imbalance is stored as debt and penalized later. We give a safety-first Pareto theory showing that the exact safe-fair constrained selector is strongly Pareto efficient for reward, safety, and fairness, with hard-safety and temporal burden-gap certificates. Empirically, the primary reported rows keep zero hard violations while improving useful burden balance: HalfCheetah-2x3-Fair improves selected safe return from $2145.5$ to $2637.4$ and Jain burden from $0.903$ to $0.977$; Safety-Gym MultiGoal-Fair reaches the lowest matched-budget scalar fairness cost with zero safety cost; and VMAS-SafeMultiGoal improves over recovery-only attribution controls. We separate social/resource fairness from mechanical workload imbalance and reserve mixed or recovery-explained rows for qualified appendix analysis.

Frontier AI models and multi-agent systems have led to significant improvements in mathematical reasoning. However, for problems requiring extended, long-horizon reasoning, existing systems continue to suffer from fundamental reliability issues: hallucination accumulation, memory fragmentation, and imbalanced reasoning-tool trade-offs. We introduce **STAR-Math**, a multi-agent framework that systematically addresses these challenges through meta-level supervision and structured Reasoner-Verifier interaction. STAR-Math is structured as an orchestrated state machine with nested challenge-step-replan loops, governed by a reasoning-free Python orchestrator that separates control from inference and bounds error propagation through trace-back and re-planning. Our key innovation is a **persistent Meta-Strategist** that maintains cross-attempt memory and exercises meta-level control by issuing high-level strategic guidance or mandatory directives, so the system can escape unproductive loops rather than stagnate or over-rely on tools. STAR-Math achieves state-of-the-art results on all eight top-tier competition benchmarks: AIME 2025-2026, MathArena Apex Shortlist, MathArena Apex 2025, Putnam 2025, IMO 2025, HMMT February 2026, and USAMO 2026. It obtains perfect scores on AIMEs, Putnam, and HMMT, and shows its largest margin on Apex 2025, scoring 93.75% compared with 80.21% by the strongest baseline GPT-5.5. Ablation studies show that the gains arise from the framework's orchestration rather than from model-level diversity since removing key components or substituting in mixed backbones consistently weakens performance.

Large language model (LLM)-powered multi-agent systems have been shown to effectively solve real-world end-to-end problems through collaboration among diverse agents. However, as LLM capabilities advance and user demands continue to grow, the range of problems that agents can address expands dramatically, while the population of available agents and tools continues to increase, giving rise to open and evolving ecosystems of autonomous agents. In this paradigm, agents are likely to become increasingly specialized, each focusing on particular capabilities or problem types. In such a setting, a central challenge is how to efficiently assemble an appropriate team of agents to take on different responsibilities from a massive agent pool for a given task. In this work, we study this problem in a controlled setting with 300 task-oriented agents. SMART (Scalable Multi-Agent Role-conditioned Teaming) first plans a task-conditioned collaboration topology and candidate role set, then constructs a compact role-conditioned candidate pool, and finally uses a LLM-free Monte Carlo Tree Search (MCTS) to assign agents to roles online via metadata-based rollouts. This design enables high flexibility without incurring prohibitive inference costs. Across multiple experiments, our approach achieves superior performance on GSM8K (93.64\%), MATH (61.73\%), HumanEval (93.89\%), and MBPP (87.68\%) using gpt-4o-mini under a limited budget.

Frontier LLMs are increasingly capable of expert-level scientific reasoning, but using them as reliable scientific agents requires more than general reasoning ability. Scientific simulation is a central test case: modern simulators are configured through executable interfaces such as XML decks, input scripts, and namelists that function as domain-specific languages tied to the simulator's internal API. Translating a researcher's natural-language intent into a runnable configuration remains a recurring expert bottleneck. We study how much of this bottleneck a general-purpose LLM coding harness, Claude Code, can absorb when wrapped with a **Simulator-Interface Grounding Adapter (SIGA)**: a package of skills, tools, and workflow control flow that grounds the agent's outputs in simulator documentation, schema, and example libraries. We instantiate SIGA for GEOS, an open-source multiphysics simulator used in CO₂-storage and induced-seismicity research. A Resolution-IV factorial surfaces three benefits over vanilla Claude Code. First, SIGA improves **reliability**, reducing across-seed variance by roughly 40× by preventing unparseable or empty decks on a hard tail of compound multi-physics tasks. Second, it improves **quality**, raising mean structural similarity by about +7 percentage points on the same hard tail. Third, a self-evolved variant matches the best hand-designed cell with roughly 16% fewer tool calls, suggesting that automatic SIGA discovery is tractable. A preliminary human baseline finds that geoscience-domain-expert volunteers new to GEOS take between 8 and 36 times as long as the agent on a representative task. An explicit human-consultation tool is used in only about 3% of under-specified trials, with the agent instead relying on the on-disk example library as a cheaper retrieval substitute. A small OpenFOAM transfer study indicates that the recipe is not specific to GEOS XML: the same stop-hook component again dominates reliability, and the best SIGA cell outperforms both vanilla Claude Code and a constrained Foam-Agent lint-only baseline. We close with SIGA-design recommendations grounded in failure modes that our best configuration still does not fix.

Cross-city transfer leverages labeled data from well-instrumented cities to improve prediction in label-scarce ones, but remains challenging when cities adopt incompatible partitions with no ground-truth region correspondences. Even with expressive GNN encoders, transfer quality varies dramatically across methods sharing nearly identical backbones---indicating that alignment design, not encoder capacity, is the binding constraint. Existing paradigms exhibit complementary failure modes: heuristic anchor matching collapses to hubness under unequal partitions, while distribution-level matching over-mixes embeddings under heterogeneity. Both stem from a single missing primitive---explicit, mass-controlled soft correspondence between unequal region sets. The challenge intensifies in multi-source transfer, where independent source-to-target alignments yield conflicting gradients and source domination. We propose SCOT, which adapts entropic OT to this regime through three application-specific designs: an OT-weighted contrastive objective that resolves the geometric--semantic tension, a one-sided cycle regularizer respecting the rectangular $n_s\!\neq\!n_t$ geometry, and---as our central contribution---a shared prototype hub coordinated through balanced entropic OT under a target-induced prior, bypassing the source-selection problem in label-scarce regimes. Across real-world cities and tasks, \scot consistently improves transfer accuracy, achieving 5--50\% relative MAE/MAPE reductions over the strongest baseline, with learned couplings and hub assignments quantitatively confirming that the diagnosed failure modes are resolved.

The exponential growth of machine learning submissions has strained the traditional peer review process, resulting in slow feedback loops for authors and an immense burden on reviewers to rigorously audit technical soundness and verify literature. To address this, we introduce ScholarPeer, a multi-agent framework designed to operationalize the rigorous auditing workflow of a senior researcher. Rather than attempting to replace human judgment, ScholarPeer serves as a co-scientist: acting as a mentor for rapid author iteration prior to submission, and as an active verification assistant that augments human reviewers. The framework structurally decouples contextualization from critique by deploying a sub-domain historian to synthesize the field's trajectory, a baseline scout to proactively hunt for omitted state-of-the-art comparisons, and a multi-aspect Q&A engine that deeply audits technical soundness—scrutinizing internal logical consistency, experimental validity, and mathematical rigor—while cross-referencing claims against top-tier academic venues. We comprehensively evaluate ScholarPeer on $\sim$1,800 ICLR submissions spanning 2020 through 2025. Our results show that ScholarPeer achieves significant win-rates against state-of-the-art fine-tuned models and search-augmented agentic baselines.

Multi-stage ML inference pipelines are difficult to autoscale because of heterogeneous resources, cross-stage coupling, and dynamic bottleneck migration. Production controllers ignore cross-stage effects, and learning-based autoscalers require offline training data that is unavailable for new pipelines. We present SAIR, an autoscaler for the zero-shot deployment regime that uses an LLM as an in-context reinforcement learning controller, improving its policy online from reward-labeled experience tuples in the LLM's context with no gradient updates. SAIR combines Pareto-dominance reward shaping with a provable separation margin between dominated and non-dominated actions, Bayesian-surprise experience retrieval, and a user-space CUDA interception layer that delivers continuous GPU rate control with sub-millisecond actuation overhead. We prove a regret bound that decomposes into retrieval coverage and LLM selection components, each mapping to a concrete SAIR mechanism. On four ML serving pipelines under three workload patterns, SAIR achieves the best or tied-best P99 latency in every configuration (up to 50\% reduction over the next baseline); billable cost is reduced through replica consolidation, and effective cost (which captures GPU rate-control savings under co-location or MIG) by up to 97\%. SAIR reaches 86\% bottleneck detection accuracy without any offline training.

Compositional implicit surface representations model scenes as collections of objects, each encoded by a Signed Distance Field (SDF). A fundamental limitation of this approach is that multiple SDFs can produce geometries that interpenetrate, violating physical plausibility. Existing mitigation strategies rely on soft penalty terms that reduce but do not eliminate intersections, and require careful loss weighting. To truly prevent interpenetration, we propose a hard constraint on vector-valued SDFs and introduce S2MDF, a lightweight plug-and-play module that enforces the constraint on any object-compositional SDF representation without architectural modifications. It introduces \fs{negligible} computational overhead and is compatible with linearly-interpolated standard meshing algorithms such as Marching Cubes. It can be applied during training or as a post-processing step. Experiments on multiple state-of-the-art compositional methods show that S2MDF reduces intersections to numerical precision while preserving reconstruction quality, outperforming existing mitigation strategies.

AI agents are governed by hundreds of natural language criteria---yet existing verifiers cannot efficiently evaluate an agent's output against a large set of such criteria and report which are satisfied, violated, or inapplicable. We introduce the VFM, a tree-attention verifier that encodes the shared context and target text, and scores $N$ natural language criteria in parallel, returning independent per-criterion ternary verdicts. The VFM is pretrained on CriteriaBank, an open corpus of 357K (context, target, criterion, verdict) tuples spanning over 285K distinct criterion strings. A finetuned VFM-4B reaches 88.3\% accuracy on a synthetic governance benchmark, surpassing larger generative judges, and on real insurance-compliance calls it is competitive with GPT-5.4 with synthetic finetuning alone. On the DynaBench multi-criterion benchmark, a finetuned VFM-4B reaches $0.85$ F1 on trace-level failure detection versus DynaGuard-4B's $0.72$; the VFM additionally localizes the violated rule as its top-1 prediction in $98.4\%$ of failing traces, and a top-3 cascade to a 26B Gemma 4 chain-of-thought judge reaches $0.94$ F1. CriteriaBank pretraining improves cross-domain transfer on four additional benchmarks, particularly in the low-data regime. On an H100, VFM-4B evaluates 1000 criteria against a 4K-token context in under 2s at $<$10GB VRAM, making it suitable for runtime verification. We will release CriteriaBank, trained checkpoints, and evaluation code.

Game balancing is a longstanding challenge requiring repeated playtesting, expert intuition, and extensive manual tuning. We introduce RuleSmith, a framework for automated game balancing that couples multi-agent LLM self-play with Bayesian optimization over a parameterized rule space. As a proof of concept, we build CivMini, a simplified civilization-style game with two asymmetric factions governed by 12 tunable parameters. LLM agents read textual rulebooks and game states to generate legal actions, enabling fast evaluation of balance metrics such as win-rate disparity. To search the rule space efficiently, we use Bayesian optimization with acquisition-based adaptive sampling: candidates with high Expected Improvement receive more evaluation games, while exploratory candidates receive fewer. RuleSmith consistently finds near-balanced configurations and produces interpretable parameter adjustments. It provides a scalable alternative to manual playtesting. Code will be available upon acceptance.

Real-world multi-view data always suffer from imperfect information problem, where the view-specific observations are absent (\ie, Incomplete Views, IV) and cross-view correspondences are mismatched (\ie, Noisy Correspondences, NC) for certain instances. As a remedy, numerous IV- and NC-oriented multi-view clustering (MvC) methods have been proposed, which however require either reliable correspondences or sufficiently complete instances, thus stopping short of addressing the imperfect information problem. In contrast, we observe that both IV and NC challenges originate from the same issue of imperfect cross-view counterpart information, where the counterpart of an anchor instance in another view might be either unavailable or unreliable. Based on the observation, we propose a novel robust MvC framework, termed Posterior-guided Latent Counterpart Inference (PLCI), which could handle both IV and NC in a unified manner. Specifically, PLCI formulates the desired cross-view counterpart of each anchor instance as a latent variable, and integrates both instance-level reliability and prototype-level semantic transport to infer the posterior distribution of the latent counterpart. Extensive experiments on six widely-used multi-view datasets against 10 state-of-the-art MvC methods demonstrate the effectiveness of PLCI for tackling the imperfect information problem. The code will be released upon acceptance.