NeurIPS 2026 论文列表

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.

Accurate RF odometry remains difficult due to multipath interference, geometric degeneracy, and dynamic clutter. Existing approaches mitigate these challenges by relying on auxiliary sensors or ground-truth pose supervision, which limits scalability and deployability. We present RFWalk, a self-supervised, RF-only odometry framework that learns cross-frame correspondence without any pose annotations or auxiliary sensors. Our key insight is to cast each range-azimuth cell as a node in a space-time graph and recover inter-frame motion by learning a soft transition matrix through cycle-consistent contrastive random walks. To resolve geometric degeneracy in feature-poor scenes, we further introduce a physics-informed Doppler consistency loss that enforces agreement between the correspondence-induced displacement field and measured radial velocities. Experiments on challenging scenes show that RFWalk achieves competitive accuracy in static environments and substantially outperforms existing approaches in dynamic scenes.

At the heart of existing language model agents is a fixed orchestrator program which is responsible for the state transition between consecutive turns. This paper introduces \emph{self-programmed execution} (SPE), an agent architecture in which the model completion is itself the orchestrator program, and the harness evaluates this program but does not impose its own orchestration policy. I formalize this idea using agentic machines: an SPE state is one from which a model completion can load any state of an embedded copy of the machine, meaning that it is subject to no fixed turn-to-turn orchestration policy. Realizing SPE in practice is nontrivial because the same data is both model context and executable program. I therefore introduce \textsc{Spell}, a Lisp-based language in which programs can edit and re-evaluate themselves, and effectful expressions like model invocations are structured such that re-evaluating an edited program does not replay its side effects. Experiments with existing models, not trained for SPE or \textsc{Spell}, show that frontier models can operate in this regime and accomplish challenging agentic tasks. These results demonstrate how an LM can act as an agent without any fixed orchestration policy, and they raise the question of what self-orchestration strategies might be learned by a model trained for self-programmed execution.

On-policy distillation (OPD) has become a popular training paradigm in the LLM community. This paradigm selects a larger model as the teacher to provide dense, fine-grained signals for each sampled trajectory, in contrast to reinforcement learning with verifiable rewards (RLVR), which only obtains sparse signals from verifiable outcomes in the environment. Recently, the community has explored on-policy self-distillation (OPSD), where the same model serves as both teacher and student, with the teacher receiving additional privileged information such as reference answers to enable self-evolution. This paper demonstrates that learning signals solely derived from the privileged teacher result in severe information leakage and unstable long-term training. Accordingly, we identify the optimal niche for self-distillation and propose \textbf{RLSD} (\textbf{RL}VR with \textbf{S}elf-\textbf{D}istillation). Specifically, we leverage self-distillation to obtain token-level policy differences for determining fine-grained update magnitudes, while continuing to use RLVR to derive reliable update directions from environmental feedback (e.g., response correctness). This enables RLSD to simultaneously harness the strengths of both RLVR and OPSD, achieving a higher convergence ceiling and superior training stability.

The landscape of self-supervised learning (SSL) is currently dominated by generative approaches (e.g. MAE) that reconstruct raw low-level data, and predictive approaches (e.g. I-JEPA) that predict high-level abstract embeddings. While generative methods are stable due to their reliable training targets based on ground-truth data, they are computationally inefficient for high-redundancy modalities like imagery, and their training objective does not prioritize learning high-level, conceptual features. Conversely, predictive methods often suffer from training instability due to their reliance on the non-stationary targets of final-layer self-distillation. We introduce Bootleg, a method that bridges this divide by tasking the model with predicting latent representations from multiple hidden layers of a teacher. This hierarchical objective forces the model to capture features at varying levels of abstraction simultaneously. We demonstrate Bootleg significantly outperforms comparable baselines (+10% vs. I-JEPA) on frozen probe classification of ImageNet-1K, iNaturalist-21, and VTAB, and semantic segmentation of ADE20K, Cityscapes, and COCO-Stuff.

Long-form reasoning and tool calling traces accumulate errors and stale content that anchor subsequent generations, a phenomenon known as \emph{context rot}. Existing scaffolds mitigate this with \textit{fixed-interval} compaction triggered at a token threshold, but such triggers are blind to trajectory structure and risk discarding partial results mid-derivation or mid-search. We propose SelfCompact, which pairs two inference-time elements: an inline \emph{compaction tool} the model invokes itself, and a lightweight \emph{rubric} specifying when to fire (a sub-task has resolved, or the trajectory is converging) and when to suppress (mid-derivation, or when stuck). Both are needed. The tool alone is unevenly used across open-weight models, often invoked at unhelpful moments or not at all; the rubric alone cannot act. Together, they elicit effective adaptive compaction without any fine-tuning. On competition math (IMO-Answerbench, HMMT Nov 25 / Feb 26) with four Qwen3 / Qwen3.5 models and agentic search (BrowseComp, BrowseComp-Plus) with three deployed agents, \method{} matches or exceeds fixed-interval summarization at a fraction of the token cost, improving over a no-summarization baseline by up to 16.7 points on math and 5--9 points on agentic search at 30--70\% lower per-question cost. The result reframes \emph{when to compact} as a meta-cognitive capability that scaffolds, not weights, can supply.

Developing generalist GUI agents capable of robust long-horizon execution remains a key challenge. While Reinforcement Learning (RL) offers a promising path for improvement, it is severely bottlenecked by the lack of reliable reward signals in open-ended environments. Ground-truth verification typically relies on expensive human annotation, which is prohibitively scarce at the scale required for training, while existing VLM-based judges frequently hallucinate success on superficial visual cues. This scarcity of trustworthy supervision fundamentally limits the scalability of current methods. In this work, we propose I-Judge, a novel framework for Self-Calibrated Reward Modeling. We leverage Inverse Dynamic Modeling (IDM) as a self-supervised training objective to learn the causal dynamics of GUI interactions from massive unlabeled trajectories, effectively bypassing the bottleneck of scarce human labels. We then introduce a runtime calibration mechanism that weights reward signals by the IDM's action consistency, filtering out spurious successes. Extensive end-to-end RL experiments on OSWorld, ScienceBoard and AndroidWorld demonstrate that our method significantly accelerates convergence and improves final agent performance. Notably, I-Judge demonstrates strong generalization to unseen domains. All the code, data and models will be made publicly available to foster further research.

Although the mechanics of eigenvalue bias of self-supervised learning (SSL) in linear networks are well-understood, they remain theoretically disconnected from empirical shortcut learning, offering little guidance or intuition. In this paper, we present a theoretical analysis of this shortcut learning phenomenon through the lens of \textit{extent bias} and \textit{amplitude bias}, two forms of \textit{eigenvalue bias}. By investigating the relations among extent bias, amplitude bias, and learning priorities in SSL, we demonstrate that SSL prioritizes features based on dimensionality (object to image ratio) and amplitude (backdoor attack intensity), not their semantic importance. Our analysis reveals how the eigenvalues of the feature cross-correlation matrix influence which features are learned earlier, providing insights into why models preferentially learn shortcut features over more generalizable features.

We present SS3D, a web-scale SfM-based self-supervision pretraining pipeline for feed-forward 3D estimation from monocular video. Our model jointly predicts depth, ego-motion, and intrinsics in a single forward pass and is trained/evaluated as a coherent end-to-end 3D estimator. To stabilize joint learning, we use an intrinsics-first two-stage schedule and a unified single-checkpoint evaluation protocol. Scaling SfM self-supervision to unconstrained web video is challenging due to weak multi-view observability and strong corpus heterogeneity; we address these with a multi-view signal proxy (MVS) used for filtering and curriculum sampling, and with expert training distilled into a single student. Pretraining on YouTube-8M (~ 100M frames after filtering) yields strong cross-domain zero-shot transfer and improved fine-tuning performance over prior self-supervised baselines. We release the pretrained checkpoint and code.

Self-speculative decoding offers a plug-and-play route to accelerating autoregressive large language model (LLM) inference, but its practical success hinges on identifying effective layer-skipping configurations online without retraining or offline profiling. Existing methods mainly optimize agreement-based proxies that capture draft-target consistency, yet overlook intrinsic runtime differences across draft configurations. Consequently, configurations with similar acceptance can still yield substantially different realized decode throughput. We present SPIRIT, a speed-driven online adaptation framework for self-speculative decoding that directly optimizes realized decode throughput, measured as generated tokens per second, using window-level runtime feedback. Our key insight is that speed is not determined by agreement alone: it is jointly shaped by draft-target consistency and the intrinsic speed advantage of the draft configuration. Empirically, the skip ratio serves as a first-order control variable, capturing the dominant consistency-speed trade-off while sharply shrinking the search space over layer-skipping patterns. SPIRIT therefore first identifies an effective skip ratio and then refines the layer-skipping pattern within it. Combined with length-aware normalization and shift-aware profile-based adaptation, this design enables rapid adaptation to task shifts in non-stationary request streams. Across diverse tasks and model scales, SPIRIT consistently outperforms strong training-free baselines and achieves up to 1.74x speedup while preserving the target model's output distribution, establishing a practical foundation for plug-and-play, high-throughput LLM inference.

A fundamental open question in self-supervised learning (SSL) is the explicit characterization of the optimal geometry of the learned representations. Recently, LeJEPA identified isotropic Gaussian embeddings as optimal for minimizing downstream prediction risk in Euclidean spaces. However, the corresponding problem for distributions supported on lower-dimensional manifolds, such as the hypersphere, remains unexplored. In this work, we demonstrate that extending this minimax analysis to smooth distributions on Riemannian manifolds fundamentally changes the optimal solution. We show that, under a worst-case formulation, both \(k\)-nearest neighbors and kernel ridge regression induce hyperspherical uniformity. More precisely, we show that uniform distributions on manifolds are optimal for \(k\)-nearest neighbors, and that the uniform distribution on the sphere is optimal for kernel ridge regression with both the exponential dot-product kernel and the linear kernel. This theoretical insight reveals a fundamental limitation of Gaussian embeddings: their non-uniform density induces anisotropic $k$-NN neighborhoods, severely biasing the estimator. To correct this, we introduce \textbf{SPHERE-JEPA}, a theoretically grounded SSL framework. We adapt LeJEPA's Cramér--Wold projection mechanism to enforce hyperspherical uniformity rather than a Gaussian prior. Empirically, SPHERE-JEPA yields significant improvements, boosting texture retrieval mAP by over 6\%, while consistently matching or outperforming LeJEPA on standard benchmarks—including a $+1.8\%$ linear probing gain on ImageNet-1K (ViT-B/14).

We identify a theoretical incompatibility between 3D point cloud masked autoencoding and SE(3)-equivariant networks. We prove that standard masking triggers a topological collapse, effectively forcing the model to hallucinate orientation without a reference frame. To resolve this, we introduce SIEVE (Scalar Invariant Extraction Via Equivariance), an SE(3)-equivariant self-supervised framework that compresses point cloud geometry into invariant scalar embeddings. At its core is the spherical projection mask, which preserves global pose while obfuscating local semantics, sidestepping the obstruction by construction. Experiments confirm that SIEVE circumvents the predicted collapse where constant and noise masks fail. On downstream point cloud tasks, the resulting embeddings improve over coordinate-based baselines and hand-crafted geometric descriptors, with the largest gains on generalization to unseen shape categories.

Machine unlearning for large language models (LLMs) aims to selectively remove memorized content such as private data, copyrighted text, or hazardous knowledge, without costly full retraining. Most existing methods require a retain set of curated examples to prevent catastrophic degradation of general model utility, creating an extra data dependency that complicates deployment. We propose SHRED (Self-distillation via High-surprisal-only Retain-set-free Entropy Demotion), a retain-set-free unlearning method built on a key insight: not all tokens within a forget set instance carry memorized information equally. High-information tokens concentrate the model's memorized knowledge, while low-information tokens reflect general language competence. SHRED operates in two stages. (1) Selection: We perform a forward pass on a forget set instance, collect per-token autoregressive probabilities, and select the bottom-$P$ (lowest probability, highest Shannon information) as forget positions; the remaining positions are retained as benign anchors. (2) Training: We construct modified KL targets that demote the memorized token's logit at forget positions while preserving the original distribution at benign positions. The model is then trained via a single top-$K$ KL self-distillation objective that simultaneously drives forgetting and utility preservation. We evaluate SHRED across four standard unlearning benchmarks and demonstrate that it establishes a new Pareto-optimal trade-off between forget efficacy and model utility, outperforming retain-set-dependent methods. Our analysis shows that SHRED is robust against relearning attacks and membership-inference attacks, and it maintains stable utility even after many sequential unlearning runs.

Black-box watermarked LLM identification has become an important task for watermark auditing. The core challenge arises from inherent black-box constraints that deny access to logits, detector keys, model parameters, and internal watermark settings. For this reason, existing methods have not conducted in-depth exploration on this task, leaving prominent limitations: **Fixed-Reference Calibration** and **Family-Specific Tests**. First, existing works use fixed null reference to serve as a baseline for judging the presence of watermarks in LLM outputs, which might misidentify a watermarked LLM as unwatermarked because the null reference is sensitive to the prompt pair, queried model, and sampling configuration. Second, existing statistical-tests based methods adopt distinct statistical features and judgment criteria for different LLM watermark families, making them less suitable when the deployed watermark algorithm is unknown. To remedy the above two limitations, this work proposes **SCTI**, a Self-Calibrated Trident Identification framework for black-box watermarked LLM identification. Specifically, to handle the first limitation, **SCTI** constructs empirical null distributions from the queried responses to estimate the generalizable reference instead of relying on a fixed reference. Meanwhile, to address the second limitation, **SCTI** is configured with a Trident-View Consistency measuring mechanism, which enables a unified pipeline to avoid designing separate tests for individual watermark family. Experimental results verify that **SCTI** outperforms representative black-box watermark identification baselines across diverse LLMs and watermark algorithms.

Self-play can train language models without external supervision. However, existing methods require rule-checkable answers, leaving open-ended tasks dependent on curated prompts or frontier-model judges. We introduce SCOPE, a data-free self-play framework for open-ended tasks that co-evolves two policies: a Challenger that generates document-grounded tasks, and a Solver that answers them through multi-turn retrieval. A frozen copy of the initial model serves as the self-judge, which writes task-specific rubrics from the source document and grades Solver responses against them. Across three 7-8B instruction-tuned models (Qwen2.5, Qwen3, OLMo-3), SCOPE improves open-ended performance by up to +10.4 points on eight benchmarks and matches or exceeds GRPO_data trained on ~9K curated prompts. Although trained only on open-ended tasks, SCOPE also improves held-out short-form QA by up to +13.8 points on seven held-out benchmarks, surpassing GRPO_data on all three models. Ablations show that co-evolving the Challenger is necessary to keep tasks near the Solver's frontier, that gains arise from improvements in both retrieval and synthesis with the relative contribution varying by task, and that rubric generation quality is the bottleneck for self-judging.

Calibrating large language models (LLMs) in open-ended generation is uniquely challenging because many distinct token sequences can express the same underlying meaning. Prior calibration techniques target either fixed-set classification confidence or the binary correctness of individual strings, making them inherently ill-suited for the dynamic, multi-string nature of open-ended semantic outcomes. To bridge this gap, we introduce S-EDL, a novel evidential method that natively optimizes LLMs for semantic calibration. By eliciting evidence from the sequence likelihoods of the model itself, S-EDL constructs a differentiable Dirichlet prior over prompt-specific semantic classes. Optimizing the evidential loss under this prior yields a calibration-aware training signal over variable semantic classes, completely bypassing the need for fixed multiple-choice options. Evaluations across four open-ended benchmarks and three models demonstrate that S-EDL substantially reduces semantic calibration error while improving accuracy. By aligning generative likelihoods with semantic correctness, this work establishes a framework for deploying trustworthy LLMs in complex, open-ended applications.

Model-assisted semi-supervised learning offers a powerful paradigm for enhancing the efficiency of statistical estimation and inference by leveraging black-box predictions on unlabeled data. However, the validity of these methods typically relies on the strict assumption that labeled and unlabeled populations share identical distributional characteristics. Consequently, state-of-the-art approaches, such as prediction-powered inference, become fragile when this assumption is violated. To address the fragility, we propose a robust framework for semi-supervised learning that remains reliable under distributional heterogeneity. By embedding a robust statistical calibrator into the prediction-rectification mechanism, the approach effectively reduces the bias arising from shifted or corrupted unlabeled samples. To maximize data utility, we introduce an adaptive cross-validation procedure to select the optimal calibrator, ensuring a reliable trade-off between statistical efficiency and robustness. Theoretical analysis confirms the consistency of the proposed estimator under mild conditions, while empirical results demonstrate its significant superiority over conventional baselines in heterogeneous environments.

Diffusion Large Language Models (DLMs) are being actively explored as promising alternatives to autoregressive models due to their fast and flexible generation capabilities. To optimize inference in DLMs, recent studies have proposed a variety of decoding algorithms. However, the tendency of these methods to mimic autoregressive generation limits the speed and flexibility of DLMs, while also preventing the methods from fully realizing their potential. Moreover, this rigid approach leads to error propagation, while existing sampling-based correction methods incur prohibitive computational costs. To overcome this, we propose Renoise Consistency, a novel post-hoc approach that enables effective self-correction by maximizing train--inference consistency and can be applied to any decoding algorithm with a flexible computational budget. Furthermore, we introduce Adaptive Block Drafting, which leverages consistent internal representations of DLMs to significantly reduce the overall computational cost. When combined, our proposed method achieves up to a 9.00\% performance improvement over the semi-autoregressive baseline, along with a 25.85\% reduction in forward steps.

We introduce ReflectDrive-2, a masked discrete diffusion planner with a separate action expert for autonomous driving that represents plans as discrete trajectory tokens and generates them through parallel masked decoding. This discrete token space enables in-place trajectory revision: AutoEdit rewrites selected tokens using the same model, without requiring an auxiliary refinement network. To train this capability, we use a two-stage procedure. First, we construct structure-aware perturbations of expert trajectories along longitudinal progress and lateral heading directions and supervise the model to recover the original expert trajectory. We then fine-tune the full decision--draft--reflect rollout with reinforcement learning (RL), assigning terminal driving reward to the final post-edit trajectory and propagating policy-gradient credit through full-rollout transitions. Full-rollout RL proves crucial for coupling drafting and editing: under supervised training alone, inference-time AutoEdit improves PDMS by at most $0.3$, whereas RL increases its gain to $1.9$. We also co-design an efficient reflective decoding stack for the decision--draft--reflect pipeline, combining shared-prefix KV reuse, Alternating Step Decode, and fused on-device unmasking. On NAVSIM, ReflectDrive-2 achieves $91.0$ PDMS with camera-only input and $94.8$ PDMS in a best-of-6 oracle setting, while running at $31.8$ ms average latency on NVIDIA Thor.