NeurIPS 2026 论文列表

Multimodal evidence can improve topic discovery, but dense auxiliary signals can make topics harder to interpret when they replace readable descriptors. This creates an evidence-role problem: symbolic signals should define what topics mean to a reader, while dense aligned signals should help infer which topics a document expresses. We introduce M3-HNTM, a hyperspherical multimodal neural topic model that separates these roles through a shared document-level latent direction. M3-HNTM utilizes a shared von Mises-Fisher (vMF) posterior for document-level topic inference and vMF-mixture symbolic decoders that represent topics as concentration-aware semantic regions. Text Bag-of-Words and speech Bag-of-Acoustic-Words provide reconstructable symbolic evidence, while aligned image-text embeddings enter posterior inference as contextual evidence and help preserve semantic structure. The model applies to image-text, speech-text, and image-speech-text corpora while keeping topics readable through symbolic descriptors. Experiments on five datasets show that M3-HNTM improves the coherence-diversity quality score over strong text-only, hyperspherical, and multimodal baselines. On SpokenCOCO-Tri, the full tri-modal model outperforms both image-text and speech-text variants, indicating that visual and acoustic signals provide complementary evidence for interpretable topic discovery.

Encoding input coordinates with sinusoidal functions into multi-layer perceptrons (MLPs) has proven effective for implicit neural representations (INRs) of surfaces defined as zero-level sets. However, existing methods often struggle to balance training efficiency, rendering speed, and noise robustness: single-MLP approaches are expensive at inference, grid-based representations are fast but can limit surface smoothness and overfit input noise, and previous multiscale approaches frequently capture noise and produce artifacts due to hard spectral truncation. To address these limitations, we propose M-plicits, a multiscale framework that models surfaces as a residual sum of MLPs trained via a sequence of nested neighborhoods. Unlike existing residual approaches that rely on standard domain-wide sampling and require costly mesh extraction for visualization, our method strictly localizes supervision to narrow bands around the previous zero-level sets. This nested design naturally provides robustness against noisy input data: the coarse network acts as a low-pass filter that establishes a clean geometric prior, while subsequent residuals progressively refine the geometry without fitting to high-frequency artifacts. We further introduce a multiscale sphere-tracing algorithm and a GEMM-based analytical normal computation that bypasses auto-differentiation entirely, yielding high-fidelity real-time rendering. On Stanford and Thingi32, M-plicits achieves the best mean Chamfer distance in the coarse configuration and the best median Chamfer distance and IoU in the fine configuration, with substantially better noise robustness than iNGP, BACON, and IDF, while using an order of magnitude fewer parameters than grid-based baselines. Code and models will be released.

We are entering a new era of *composite model architectures* that integrate diverse components such as vision encoders, language backbones, diffusion and flow heads, audio codecs, action generators, and world-model predictors. Such architectures underpin a broad class of multimodal models, including unified multimodal models, omni models, speech-language models, vision-language-action policies, and world models. However, existing model serving frameworks were built on narrow assumptions about model structure, making them ill-suited to accommodate this new architectural diversity. Here we present M*, a universal serving system for efficient serving of composite AI models. M* represents models as dataflow graphs, processing requests spanning diverse modalities and tasks as traversals over these graphs. The core insight is a modular abstraction that supports arbitrary composition of model components, flexible placement onto a physical cluster, and model-agnostic optimizations within a distributed runtime. We call this abstraction the *Walk Graph* and show how it can concisely capture composite models from a broad range of families. We instantiate M* on representative models and find that it achieves, on average, 30\% lower end-to-end latency than vLLM-Omni for text-to-image workloads on BAGEL, while delivering a lower real-time factor and higher throughput - by up to 15\% - for text-to-speech workloads on Qwen3-Omni. M* also outperforms the V-JEPA 2-AC rollout baseline for robotic planning by up to $12.5\times$. Thus, our work paves the road towards more efficient serving of complex models with minimal developer effort.

Neural surrogate models for physical simulations are trained on discretized samples of continuous domains, where the induced empirical measure leads to uneven supervision, biasing optimization and causing spatial inconsistencies in physical fidelity. To mitigate this measure-induced bias, we propose M$^3$ (Multi-scale Morton Measure), a scalable framework that balances training measures by partitioning space according to physical variation and allocating supervision across multiple scales. Applied to three industrial-scale datasets with diverse discretizations, M$^3$ consistently improves predictions in the continuous physical domain, achieving up to 4.7$\times$ lower error in large-scale volumetric cases. These gains persist under aggressive subsampling (160M $\rightarrow$ 16M $\rightarrow$ 1.6M points), where M$^3$-trained models outperform those trained on higher-resolution data, reducing physics-weighted relative $L_2$ error by 3--4$\times$ and the corresponding MSE by up to 13$\times$. These results highlight data distribution as a key factor in operator learning and position M$^3$ as a scalable, data-efficient approach for physically consistent modeling.

We study online scheduling in a shared-resource system where tasks of different types arrive sequentially and compete for limited resources. Each task is modeled as an $H$-stage episodic Markov decision process, which captures the sequential resource-allocation decisions made within the task. At the system level, the learner must coordinate resource sharing across tasks to maximize the long-term ratio of cumulative reward to cumulative cost over $K$ tasks, without knowledge of the task distribution. The key challenges are coupled intra-task Markovian dynamics and inter-task ratio maximization, further complicated by unknown models and large or continuous state spaces. To address these challenges, we propose Contextual Optimistic Lyapunov Decomposition (COLD), a Lyapunov-driven optimistic learning algorithm. COLD converts the long-term ratio objective into a queue-adjusted per-task surrogate, thereby decoupling reward-cost optimization from statistical value estimation. It uses double-optimistic least-squares value iteration to estimate reward and cost values under linear function approximation, and a softmax policy to smooth the greedy update with only a controlled approximation error. We prove a regret bound of $\tilde{\mathcal{O}}(\sqrt{Md^3H^3T})$, where $M$ is the number of task types, $d$ is the feature dimension, $H$ is the episode horizon, and $T=KH$ is the number of interaction steps. The $\sqrt{T}$ dependence is order-optimal up to logarithmic factors, and the bound scales with the feature dimension rather than the size of the state space. Experiments on synthetic environments and an empirical KuaiRand-based MDP demonstrate the effectiveness of the proposed method.

Random forests are widely used in fields involving sensitive tabular data, but existing approaches to enforcing differential privacy (DP) typically degrade performance to the point of impracticality. In this paper, we introduce Lumberjack, a differentially private random forest algorithm that achieves substantially higher utility by constructing large random decision trees and then applying aggressive, privacy-preserving pruning to retain only sufficiently populated nodes. A key component of our approach is a novel $(\varepsilon,\delta)$-DP heavy hitter detection algorithm for hierarchical data, whose error is $O_{\varepsilon,\delta}(\sqrt{\log h})$ for trees of height $h$ and may be of independent interest. This favorable scaling enables the use of significantly deeper trees than in prior work, leading to improved expressiveness under privacy constraints. Our empirical evaluation on benchmark datasets shows that Lumberjack consistently outperforms prior DP random forest methods, establishing a new state of the art. In particular, our approach yields substantial improvements in the privacy-utility trade-off for practical privacy budgets. Our findings suggest that more carefully designed DP random forests can close much of the utility gap, highlighting a promising and underexplored direction for future research.

Generative real-world image super-resolution (Real-ISR) can synthesize visually convincing details from severely degraded low-resolution (LR) inputs, yet its stochastic sampling makes a critical failure mode hard to avoid: outputs may look sharp but be unfaithful to the LR evidence, exhibiting semantic or structural hallucinations. Preference-based reinforcement learning (RL) is a natural fit because each LR input yields a rollout group of candidate restorations. However, effective alignment in Real-ISR is hindered by three coupled challenges: (i) the lack of an LR-referenced faithfulness signal that is robust to degradation yet sensitive to localized hallucinations, (ii) a rollout-group optimization bottleneck where scalarizing heterogeneous rewards before normalization compresses objective-wise contrasts and weakens DiffusionNFT-style reward-weighted updates, and (iii) limited coverage of real degradations, which restricts rollout diversity and preference signal quality. We propose LucidNFT, a multi-reward RL framework for flow-matching Real-ISR. LucidNFT introduces LucidConsistency, a degradation-invariant and hallucination-sensitive LR-referenced evaluator trained with content-consistent degradation pools and original-inpainted hard negatives; a decoupled reward normalization strategy that preserves objective-wise contrasts within each LR-conditioned rollout group before fusion; and LucidLR, a large-scale collection of real-world degraded images for robust RL fine-tuning. Extensive experiments show that LucidNFT improves perceptual quality on strong flow-based Real-ISR baselines while generally maintaining LR-referenced consistency across diverse real-world scenarios.

We show that frontier AI models exhibit loyalty capture---a form of structural sycophancy in which a reporting relationship to a manager with misaligned incentives is sufficient to bias the model's recommendations, without any explicit instruction, user feedback, or adversarial prompting. In 5,757 trials across eight models from eight AI laboratories, CEO-reporting models cut investment to boost the CEO's bonus by 8.1 percentage points more when the bonus is at stake. This same effect is also predictably absent from a metric-irrelevant placebo. Delivered rationales reveal motivated reasoning: models adopt the bonus target, calculate how their recommendation achieves it, and dismiss long-term value, while concealing their recognition of the conflict from the recommendation recipient. Designed interventions (conflict disclosure, stakeholder broadening) eliminate ~60% of the effect; passive oversight does not. Taken together, the results indicate that AI models deployed within organizational hierarchies can develop and act on loyalty to their supervisor at the expense of the organization they are meant to serve.

Long-to-Short (L2S) model merging seeks to combine the strong reasoning capability of long chain-of-thought models with the concise style of base models, but existing methods that rely on linear interpolation and first-order activation statistics fail to capture how the loss landscape responds to parameter displacement, leading to degraded accuracy on hard tasks. We propose Low-Rank Hierarchical Merging (LHM), a training-free framework that replaces heuristic statistics with lightweight second-order curvature information. LHM consists of three components: (1) a low-rank Hessian approximation that efficiently estimates per-layer Hessian traces via Hutchinson's estimator on random inputs, eliminating the need for task- specific data; (2) a Layer Second-Order Sensitivity (LSOS) index that combines Hessian trace with task-vector norm to quantify per-layer fusion difficulty; and (3) a Dual-Constraint Coefficient Mapping (DCCM) that converts LSOS scores into layer-adaptive merge coefficients while jointly satisfying a fusion-error bound and a global length-reduction target. Experiments across six mathematical benchmarks on 1.5B and 7B models show that LHM outperforms state-of-the-art methods such as ACM-TIES and ORCA, achieving over 4\% accuracy gain and up to 93.1\% reduction in output length relative to the long-CoT model.

Scaling critic capacity is a promising direction for enhancing off-policy reinforcement learning (RL). However, larger critics are prone to overfitting and unstable in replay-buffer-based bootstrap training. This paper leverages Low-Rank Adaptation (LoRA) as a structural-sparsity regularizer for off-policy critics. Our approach freezes randomly initialized base matrices and solely optimizes low-rank adapters, thereby constraining critic updates to a low-dimensional subspace. Built on top of SimbaV2, we further develop a LoRA formulation, compatible with SimbaV2, that preserves its hyperspherical normalization geometry under frozen-backbone training. We evaluate our method with SAC and FastTD3 on DeepMind Control locomotion and IsaacLab robotics benchmarks. LoRA consistently achieves lower critic loss during training and stronger policy performance. Extensive experiments demonstrate that adaptive low-rank updates provide a simple, scalable, and effective structural regularization for critic learning in off-policy RL.

Continual learning studies how models can adapt to new tasks while retaining previously acquired knowledge. Although a broad spectrum of methods has been proposed to mitigate catastrophic forgetting, the field remains predominantly performance-driven, with limited insight into what forgetting actually corresponds to within the vision model's representation space. Prior work has primarily analyzed forgetting through task-level performance or coarse measures of representational drift, without disentangling output-level accessibility from changes in finer-grained internal structure.To this end, we propose a diagnostic framework that leverages Sparse Autoencoders (SAEs) to define a task-anchored latent feature space, enabling analysis of how task-specific information evolves at a finer granularity, where individual SAE latents are treated as concept proxies for recurring and relatively disentangled visual patterns in the model’s internal computations. Within this framework, we decompose forgetting into apparent concept deletion, recoverability, and decodability. We show that a large portion of seemingly lost concept-level information can often be recovered under linearity assumption, with concept decodability degrading as more tasks are introduced. Overall, our findings suggest that a significant part of concept-level forgetting can be attributed to changes in the representational accessibility rather than complete information erasure.

Embodied Spatial Reasoning (ESR) is central to deploying Vision-Language Models (VLMs) in real-world embodied tasks, yet remains inadequately evaluated. Existing benchmarks primarily assess spatial reasoning in static images or synthetic indoor environments, where semantic shortcuts often suffice, making it difficult to faithfully evaluate spatial reasoning capabilities in the wild. In light of this, we present Lost on Campus, a benchmark for evaluating ESR in large-scale real-world outdoor 3D environments reconstructed by 3D Gaussian Splatting. Our benchmark introduces a unified reasoning-action evaluation framework that seamlessly integrates diagnostic QA for isolated reasoning with closed-loop interactive navigation for active reasoning under multimodal instructions. To enable fine-grained diagnosis, we systematically decompose ESR into six fundamental capabilities: action grounding, spatial foresight, metric awareness, goal-directed planning, global localization, and spatio-temporal consistency. Extensive experiments reveal that: (i) Compared to indoor settings, real-world outdoor environments impose substantially higher demands on spatial reasoning, where existing models exhibit poor performance; (ii) VLMs still struggle with fine-grained visual alignment, with failures in spatial foresight and self-aware localization emerging as primary bottlenecks; (iii) Improving multimodal interaction capabilities and long-term spatial reasoning is crucial for advancing embodied intelligence. These findings highlight that faithful evaluation of ESR demands benchmarks tightly coupling perception, reasoning and action within realistic and continuous environments.

Our aim is to develop a unified model for sign language understanding that performs both sign language translation (SLT) and sign–subtitle alignment (SSA). Together, these two tasks enable the conversion of continuous signing videos into spoken language text and also the temporal alignment of signing with subtitles -- both beneficial for practical communication, large-scale corpus construction, and educational applications. To achieve this, our approach is built upon three design choices: (i) a lightweight visual backbone that captures manual and non-manual cues from human keypoints and lip-region images while preserving signer privacy and enabling end-to-end training; (ii) a Sliding Perceiver mapping network that aggregates consecutive visual features into word-level embeddings to bridge the vision–text gap; and (iii) scaling training to large datasets -- BOBSL covering British Sign Language (BSL) and YouTube-SL-25 covering American Sign Language (ASL) -- to promote generalisation across languages and signers. With this multi-task, multilingual pretraining and strong model design, we achieve state-of-the-art results on BOBSL, How2Sign, Phoenix14T, and FLEURS-ASL SLT benchmarks, and also for SSA on BOBSL and WMT-SLT-SRF. Beyond these standard, signer-interpreted datasets, we show successful SLT on the natural signing data from the BSL-Corpus.

Gradient-based machine unlearning methods typically ``unlearn'' by optimizing forget and retain losses, but similar losses do not necessarily imply similar output-space behavior. Prior work has shown that high forget loss alone does not certify that unlearning has produced retraining-like output behavior. We therefore ask: when a gradient-based unlearner differs from the model that would have been learned on only the retain data, what structure does the resulting output-space gap have? We derive closed-form expressions for this gap in linear regression, with an extension to nonlinear models via a generalized Gauss-Newton and local linearity approximations. Our analysis puts various gradient-based methods under a unified framework parametrized by geometry, reweighting, and projection. Two leverage matrices emerge that capture the forget-retain entanglement under the retrained model geometry: forget self-leverage and retain cross-leverage. We show both theoretically and empirically that these two leverage matrices drive the impact of the forget residual on the forget and retain outputs, and existing methods account for this geometry only partially. Our framework yields precomputable leverage diagnostics for auditing forget sets to identify what type of gradient-based unlearning is likely to be successful.

Transformers have achieved remarkable success across a wide range of applications, and a growing body of work suggests that part of their strength comes from their ability to learn and execute algorithmic procedures. However, our understanding of how transformers learn such algorithms remains limited, especially in the presence of layer normalization (LN). In this work, we study principal component prediction as a concrete testbed for understanding the training dynamics of transformers with LN. We prove that a looped linear transformer with LN, trained by gradient descent, converges to a solution that implements the power method, with each self-attention layer performing one power iteration. Notably, the model is trained only for principal component prediction, rather than being explicitly supervised to implement the power method. Our finding thus reveals an ``algorithmic implicit bias'' of looped transformers with LN: principal-component prediction can in principle be achieved by many mechanisms, yet gradient descent selects one that realizes the power method. We further provide a concrete comparison between transformers with and without LN: even with layerwise guidance from power iterations, a transformer without LN cannot exactly learn the power method, whereas the corresponding transformer with LN can, leading to a provable performance gap in principal component prediction. Our results provide, to our knowledge, the first theoretical analysis of the training dynamics of looped and single-layer transformers with LN, and shed light on the role of LN in transformer models.

Masked diffusion models (MDMs) have emerged as a promising alternative to autoregressive models for language modeling, yet the effective design of transformer architectures for MDMs remains underexplored. In this paper, we show that selectively looping the early-middle transformer layers significantly improves both training efficiency and model performance in MDMs. We call this approach \textbf{LoopMDM} (Looped Masked Diffusion Model), which brings two key benefits: looping layers at training-time yields a depth-scaling effect without adding parameters, while varying the number of loops at inference-time enables flexible compute scaling. Despite the simplicity, the results are striking: across multiple pre-training corpora, LoopMDM matches the performance of same-size MDMs with up to $3.3\times$ fewer training FLOPs, while its final performance outperforms them on various reasoning benchmarks, including up to $+8.5$ points on GSM8K. It even surpasses deeper non-looped MDMs trained with comparable per-step compute, indicating that selective looping is more effective than naive depth scaling. Furthermore, LoopMDM can scale inference-time compute by increasing the number of loops. Adaptively adjusting the number of loops throughout the sampling process further yields additional gains in compute efficiency while maintaining performance. Lastly, with attention analysis, we provide evidence that looping is effective in MDMs by promoting interactions among masked positions. Our code and weights will be publicly released.

Large Language Models face a challenging multi-objective coupled optimization problem in constrained long-text generation, requiring the simultaneous balancing of global structure, local coherence, and constraint satisfaction. However, existing methods often rely on static pipelines or loosely coupled designs, making it difficult to achieve cross-stage collaborative optimization. To address this, we propose LoopWeaver, which models the generation process as a hierarchical, feedback-driven closed-loop optimization problem; by introducing iterative feedback between the planning and generation stages, it achieves the joint optimization of both global and local aspects. Specifically, LoopWeaver integrates constraint-aware hierarchical planning, feasibility filtering, and reward-based preference optimization, thereby allowing feedback signals to permeate both the planning and generation layers. Unlike methods that utilize feedback solely during post-processing or within a single module, LoopWeaver introduces a unified paradigm for feedback-coupled optimization, demonstrating significant improvements in text quality and constraint satisfaction capabilities across various models.

Looped language models (LoopLMs) perform iterative latent computation to refine internal representations, offering a promising alternative to explicit chain-of-thought (CoT) reasoning. However, existing reinforcement learning (RL) paradigms primarily target output tokens, creating a structural mismatch with looped architectures whose reasoning unfolds implicitly. In this work, we propose LoopRPT, a reinforcement pre-training framework tailored for LoopLMs. By reframing next-token prediction as a next-token reasoning task, LoopRPT assigns reinforcement signals directly to latent steps using an EMA teacher reference and noisy latent rollouts. This formulation enables RL to directly shape intermediate representations, compressing effective reasoning into fewer iterations. We instantiate LoopRPT on the Ouro architecture across multiple model scales. Results demonstrate that LoopRPT consistently improves per-step representation quality, achieving Pareto dominance in accuracy–computation trade-offs. Notably, significant gains on hard tokens indicate that LoopRPT enhances early-stage reasoning rather than merely encouraging premature exits. Our findings highlight reinforcement pre-training as a principled paradigm for learning efficient latent reasoning in LoopLMs.

Structured pruning offers hardware-agnostic acceleration for Large Language Models (LLMs) but typically degrades performance by strictly adhering to the original static sequential execution graph. In this paper, we challenge this rigidity with LoopPrune, an evolutionary framework that transforms model compression into an architectural re-discovery process. By treating pre-trained layers as a library of reusable primitives, LoopPrune optimizes a dynamic execution path where modules can be skipped, executed once, or iterated via controlled loops. This approach effectively decouples effective model depth from parameter count, allowing compressed models to maintain high reasoning capacity. To navigate the vast combinatorial search space, we introduce a hybrid evolutionary strategy enhanced by a Multi-Tier Population Initialization mechanism, which stratifies the search start-point using cross-model elite transfer, structural priors, and Wanda score importance estimation. Extensive experiments on diverse LLM families demonstrate that LoopPrune sets a new state-of-the-art across all sparsity levels. Notably, LoopPrune identifies efficient iterative patterns that allow compressed models to match or even surpass the zero-shot reasoning accuracy of their dense counterparts on specific benchmarks, while maintaining near-lossless perplexity at high retention rates.