NeurIPS 2026 论文列表

Image tokenizers play a central role in modern generative models, where the structure of the latent space critically determines the downstream generation performance. A key but underexplored property of effective latent representations is spectral organization: the ability to encode information across frequency components. In this work, we introduce structured state-space regularization, a principled approach for inducing spectral structure in latent spaces. We derive a regularization objective by revisiting state-space models (SSMs) as dynamical systems over compressed representations. This perspective reveals that hidden states of SSMs follow basis function dynamics under predefined input transformations, resulting in a novel regularizer form that enforces the latent space to capture spectral components of images. Experiments demonstrate that our regularizer improves the generative performance of image tokenizers while incurring only minimal loss in their reconstruction fidelity.

As autonomous coding agents become capable of handling increasingly long-horizon tasks, they have gradually demonstrated the potential to complete end-to-end software development. Although existing benchmarks have recently evolved from localized code editing to from-scratch project generation, they remain confined to structurally simplified, single-stack applications. Consequently, they fail to capture the heterogeneous environments, full-stack orchestration, and system-level complexity of real enterprise Software as a Service (SaaS) systems, leaving a critical gap in assessing agents under realistic engineering constraints. To fill this gap, we introduce SaaSBench, the first benchmark designed to explore the boundaries of AI agents in enterprise SaaS engineering. Spanning 30 complex tasks across 6 SaaS domains with 5,370 validation nodes, it incorporates 8 programming languages, 6 databases, and 13 frameworks to meticulously mirror real-world software heterogeneity. Furthermore, we design a dependency-aware hybrid evaluation paradigm tailored for complex systems with long horizons and multi-component coupling, enabling fine-grained, reproducible assessment. Crucially, our extensive experiments reveal a striking insight: the primary bottleneck for state-of-the-art agents is not generating isolated code logic, but successfully configuring and integrating a multi-component system. Over 95\% of task failures occur before agents even reach deep business logic, with models often falling victim to overconfidence and prematurely halting during foundational system setup, or getting trapped in ineffective debugging loops. We hope SaaSBench serves as a practical and challenging testbed to drive the evolution of reliable, system-level coding agents.

Flow matching has recently emerged as a powerful framework for continuous-time generative modeling. However, when applied to long-tailed distributions, standard flow matching frameworks are susceptible to *majority bias*, over-representing dominant modes at the expense of tail-class fidelity and distributional accuracy. In this work, we propose ***Unbalanced Optimal Transport Reweighted Flow Matching (UOT-RFM)***, a novel framework for generative modeling under class-imbalanced (long-tailed) distributions that operates without any class label information. Our method constructs the conditional vector field using mini-batch Unbalanced Optimal Transport (UOT) and mitigates majority bias through a principled inverse reweighting strategy. The reweighting relies on a *label-free majority score*, defined as the density ratio between the target distribution and the UOT marginal. This score quantifies the degree of majority based on the geometric structure of the data, without requiring class labels. By incorporating this score into the training objective, UOT-RFM theoretically recovers the target distribution with first-order correction ($k=1$) and empirically improves tail-class generation through higher-order corrections ($k > 1$). Our model outperforms existing flow matching baselines on long-tailed benchmarks, while maintaining competitive performance on balanced datasets.

Simultaneous Sign Language Translation (SLT) is critical for real-time communication, yet existing methods remain largely confined to sentence-level, offline settings that assume pre-segmented inputs. These assumptions hinder deployment in realistic scenarios involving continuous, unsegmented video streams. We present ReSCUE, a unified framework for simultaneous SLT on unsegmented long-form sign language videos that aligns training and inference with realistic streaming conditions. ReSCUE combines inference-aware training to handle partial inputs, non-signing pauses, and multi-sentence contexts, stabilized re-translation to enable low-latency yet revisable predictions with reduced output flicker, and a sentence commitment mechanism for online segmentation and memory management. Experiments on standard sentence-level benchmarks show that ReSCUE achieves lower latency and the best translation quality under low-latency settings. On long-form unsegmented datasets, ReSCUE approaches the translation quality of oracle offline systems that use ground-truth sentence boundaries, while operating at substantially lower latency, demonstrating its practicality for real-world streaming scenarios.

Token-level maximum likelihood and mean negative log-likelihood dominate the training and evaluation of large language models (LLMs), yet long-horizon generation often suffers from late-stage degradation despite improved mean loss. We identify two coupled causes of this mismatch: mean token loss is not scale-free with respect to sequence length, so small per-token divergences can amplify into large sequence-level distribution shifts, and long-horizon failures are spike-dominated, where rare but abrupt prefix deviations trigger errors that are largely invisible to mean objectives. To address these issues, we propose \textbf{Prefix Likelihood-Ratio Control} (PLRC), which centers training on the prefix log-likelihood ratio process and its increments to directly capture sequence-level deviation and spikes. As the data likelihood is unknown, PLRC learns a prefix ratio critic via noise-contrastive prefix discrimination and introduces tail-sensitive regularizers on the terminal ratio and its increments, yielding guarantees that bound long-horizon event distortion and the probability of the first spike. PLRC preserves standard teacher-forcing training without architectural changes while explicitly targeting failure modes that mean token loss cannot capture.

Transformer-based video generative models have become increasingly realistic, but long-horizon consistency remains challenging to achieve because even a few dozen frames create impractically long sequence lengths. We show that this issue can be mitigated by generating video using coarse-to-fine rollout within a multi-scale token space. Our approach is simple: first, we pre-train an adaptive autoencoder that compresses each frame into a hierarchy of tokens, with levels ranging from the typical latent resolution to only a handful of tokens per frame. This yields a hierarchy in which the coarsest levels capture the most consequential information—such as scene layout and semantics—while finer levels add high-frequency appearance and texture. Then, we train a video diffusion model to generate these tokens using coarse-to-fine rollout. By carefully controlling the level of detail at which frames are generated and used as context during each rollout step, we are able to preserve long-range consistency in geometry and object permanence while expending less compute on long-term consistency of less perceptually relevant details. We validate this approach using a custom long-horizon Minecraft dataset, where it produces substantially more consistent rollouts compared to strong baselines. We hope this perspective opens new opportunities for long-term consistent video generation.

Spatially consistent long-horizon video generation aims to maintain temporal and spatial consistency along predefined camera trajectories. Existing methods mostly entangle memory modeling with video generation, leading to inconsistent content during scene revisits and diminished generative capacity when exploring novel regions, even trained on extensive annotated data. To address these limitations, we propose a decoupled framework that separates memory conditioning from generation. Our approach significantly reduces training costs while simultaneously enhancing spatial consistency and preserving the generative capacity for novel scene exploration. Specifically, we employ a lightweight, independent memory branch to learn precise spatial consistency from historical observation. We first introduce a hybrid memory representation to capture complementary temporal and spatial cues from generated frames, then leverage a per-frame cross-attention mechanism to ensure each frame is conditioned exclusively on the most spatially relevant historical information, which is injected into the generative model to ensure spatial consistency. When generating new scenes, a camera-aware gating mechanism is proposed to mediate the interaction between memory and generation modules, enabling memory conditioning only when meaningful historical references exist. Compared with the existing method, our method is highly data-efficient, yet the experiments demonstrate that our approach achieves state-of-the-art performance in terms of both visual quality and spatial consistency.

The key--value (KV) cache is a major bottleneck in long-context inference, where memory and computation grow with sequence length. Existing KV eviction methods reduce this cost but typically degrade performance relative to full-cache inference. Our key insight is that full-cache attention is not always optimal: in long contexts, irrelevant tokens can dilute attention away from useful evidence, so selective, learnable eviction can improve generation rather than merely approximate the full cache. We introduce a global retention-based KV eviction method that learns each token's future utility under a unified memory budget. Lightweight retention gates assign utility scores to cached KV entries, and a shared final scoring projection calibrates these scores across all layers and heads. This enables a single global eviction policy in which tokens from different layers, heads, and modalities compete directly for cache capacity. We further provide theoretical analysis showing that preferentially retaining useful tokens reduces attention dilution, and we justify geometric retention as a query-agnostic proxy for future utility. Across long-context language and vision--language benchmarks, our method substantially reduces KV memory while matching or surpassing full-cache inference. These results suggest that learned, globally calibrated KV eviction is not only a compression technique, but also a mechanism for improving long-context reasoning.

Current datasets for text-to-video generation are largely limited to short clips depicting a single event or coarse-grained captions. As a result, they provide limited supervision for modeling long-take temporal dependencies, often leading to temporally inconsistent generations when describing multiple events. In this paper, we introduce MEV, a multi-event video dataset designed for long-take video generation. MEV provides high-quality real-world videos where subjects can perform a sequence of distinct, non-overlapping events, featuring precise temporal boundaries and fine-grained procedural annotations. This event-centric discretization reduces confounding factors and enables models to support temporally localized conditioning and build long-take videos from atomic event units. Furthermore, we introduce a comprehensive annotation pipeline that segments videos into temporally grounded events, and generates structured captions at both the global level and the event level. In addition, under an event-centric evaluation protocol, we conduct multi-event generation experiments by fine-tuning on MEV with temporally grounded event conditioning, leading to improved long-take event transitions and more coherent content. We believe this work removes a key barrier to training video foundation models for multi-event scenarios and provides practical insights into learning from temporally-structured video-text data.

Long-form generation in language models often collapses into repetitive loops well before the nominal context limit. This is commonly attributed to model capacity or training data, and addressed through costly retraining or post-training pipelines. We show that the sampling algorithm is a primary determinant of long-context output quality, improving outputs without further training costs. Across ten open-weight models from 1B to 120B parameters, output lengths from 8K to 64K tokens, and more than 13,500 generations, full-distribution-aware samplers (P-less, Top-H, top-$n\sigma$) dominate automated metrics and human evaluation, while standard truncation (top-$p$, top-$k$) collapses into late-context repetition; min-$p$ falls in a middle tier. A 1,500-annotation Amazon Mechanical Turk study finds P-less winning more than nine in ten head-to-head pairings against a top-$p$ 0.9 baseline on a 120B model, with 10-gram repetition approaching zero at 64K tokens. A per-step selected-token rank diagnostic isolates the mechanism: standard truncation drives the rank to the argmax floor by the final quartile, while full-distribution-aware methods stay in a healthy band across the full window. This gap that widens with model scale, establishing sampler choice as an under-appreciated bottleneck for long-context generation in open-weight models. We frame long-context generation as sampling from outside the training distribution, and encourage further study for usecases like long-context agentic coding.

Long-form technical text generation underpins knowledge-intensive workflows, yet remains challenging for large language models (LLMs) due to the need for globally consistent logical structuring and faithful technical reasoning beyond local coherence. Patent drafting is a canonical instance of this challenge, demanding holistic generation of a legally compliant and technically exhaustive document through sustained multi-expert collaboration. Existing approaches often focus on partial section generation or rely on manually crafted outlines, limiting scalable automation in realistic settings. In this work, we propose LogicTree-RAG, a logic tree-guided retrieval-augmented generation framework that induces a hierarchical logic tree as a global organizational backbone to organize and ground technical disclosures, without relying on expert-defined drafting priors. Each node in the logic tree represents a technical element and is constructed through evidence-guided recursive generation. A hybrid traversal mechanism then maps the logic tree into patent sections, enabling controllable and section-balanced generation. Extensive experiments show that LogicTree-RAG consistently improves content quality and language conformity over strong LLM-based baselines and achieves longer structured generation with high token efficiency, demonstrating the effectiveness of logic-centric generation for complex technical document drafting.

Autoregressive video generation has achieved impressive results on short clips, yet generating long videos remains challenging due to error accumulation over extended horizons, where each predicted frame depends on potentially imperfect previous frames. Existing approaches typically rely on static context management strategies, such as sliding-window caches that discard early frames or fixed sink frames that permanently preserve initial content. These designs either lose critical long-range context or anchor generation to outdated information, which can lead to degraded fidelity, motion stagnation, and reduced diversity. In this paper, we propose \textbf{Learning to Memorize (L2M)}, a learned context memory management framework that dynamically retains, evicts, and evolves historical frames according to predicted importance. The framework consists of an \emph{importance prediction router} that estimates the relevance of each historical frame, \emph{degradation-aware training} that improves robustness to noisy or corrupted context, and a \emph{dynamic memory initialization and evolution} mechanism that updates long-term memory using exponential moving average (EMA) importance scores. This allows high-quality frames to replace stale anchors and enables memory to adapt naturally to scene evolution. Trained with a teacher-student distillation objective and sparsity regularization, L2M makes efficient use of a fixed KV-cache budget while preserving the most informative contextual information. Extensive experiments show that L2M achieves superior long-term video generation performance over existing baselines, improving stability and visual fidelity, and establishing a new paradigm for learned memory management in long autoregressive video generation.

Autoregressive video diffusion models have demonstrated effectiveness in interactive game generation, with Minecraft gameplay serving as a representative application. To faithfully simulate gameplay, a model must generate natural content when exploring new scenes while preserving spatial consistency when revisiting explored areas. Under limited computation budgets, it must compress and exploit historical cues within a finite context window, which exposes a trade-off: models relying solely on temporal context offer flexible exploration but suffer from poor revisit consistency, whereas adding spatial memory strengthens consistency but may degrade new scene generation quality when the model over-relies on sparse or unreliable spatial context. We present GeoMemory, a learning framework that pairs training protocols with a geometry-indexed spatial memory. Specifically, our Hybrid Training exposes the model to both exploration and revisitation regimes, guiding the model to rely on temporal memory in new scenes while effectively incorporating spatial memory upon revisits. Chained Forward Training creates larger pose variations and encourages reliance on spatial memory for maintaining consistency. For spatial memory, we integrate Point-to-Frame Retrieval with an incremental 3D cache generated by VGGT, enabling constant-time retrieval of relevant historical context regardless of sequence length. Extensive experiments demonstrate that GeoMemory achieves superior performance in both long-term spatial consistency and visual quality in new scenes with real-time interaction.

Long-form spatial reasoning in vision--language models (VLMs) is often bottlenecked by autoregressive chain-of-thought (CoT) decoding, where intermediate reasoning traces and final answers are generated token by token. This sequential decoding path makes high-quality spatial reasoning expensive in latency-sensitive embodied and interactive settings. We propose \emph{Diffusion Thinking}, a post-training framework that converts pretrained autoregressive VLMs into diffusion-style reasoners for fast long-form spatial inference without changing the visual encoder or language backbone. Diffusion Thinking partitions each CoT trace into blocks and refines tokens within the current block in parallel through discrete denoising, while preserving causal dependence across blocks and reusing KV cache for streaming generation. This creates a high-speed operating point, but aggressive parallel refinement can weaken fine-grained reasoning. To address this, we introduce \emph{Diffusion Rethinking}, a cached test-time scaling mechanism that reallocates part of the saved latency budget to self-correction rounds, yielding a controllable speed--accuracy frontier. We construct a 561K-instance Long-CoT Spatial Reasoning dataset with verified final answers and generated CoT traces across six spatial task types. Across Qwen2.5-VL and InternVL3 backbones, Diffusion Thinking with block size $D=32$ achieves 38.18--41.81$\times$ effective wall-clock speedup over autoregressive Long-CoT decoding. With eight rethinking rounds, it retains 4.20--4.60$\times$ speedup and matches or surpasses autoregressive Long-CoT accuracy on the largest backbones. These results show that diffusion-style compute reallocation is a practical path toward fast, accurate long-form spatial reasoning in VLMs.

Autoregressive neural simulators now match classical solvers on short-horizon prediction of physical systems, yet their accuracy degrades rapidly when rolled out over long horizons. In this work we identify *transient amplification* of perturbations around rollout trajectories as a structural mechanism driving rollout error. Using a linearization analysis we show that when the Jacobians along an autoregressive trajectory are non-normal and non-commuting, the model amplifies errors transiently, resulting in model rollout drift even when the overall system is asymptotically stable. Building on the analysis, we propose \emph{commutativity regularization}: a combination of two penalties designed to reduce the normality defect of individual Jacobians and the commutator norm of Jacobians across steps. The penalties are estimated with Jacobian-vector products and have no inference-time cost. We show a propagator bound that quantifies rollout error under approximate commutativity and normality. We evaluate UNet and FNO variants with commutativity regularization on 1D and 2D spatio-temporal data in synthetic and real settings, showing successful long-horizon rollouts over thousands of steps. We show that the method improves FourCastNet climate forecasts on ERA5 without using any new data. The gain is most pronounced out-of-distribution: trained on trajectories of a few hundred steps, regularized models remain in-distribution for thousands of rollout steps on initial conditions where baselines diverge.

Autoregressive video generation aims at real-time, open-ended synthesis. Yet, cinematic storytelling is not merely the endless extension of a single scene; it requires progressing through evolving events, viewpoint shifts, and discrete shot boundaries. Existing autoregressive models often struggle in this setting. Trained primarily for short-horizon continuation, they treat long sequences as extended single shots, inevitably suffering from motion stagnation and semantic drift during long rollouts. To bridge this gap, we introduce \textbf{CausalCine}, an interactive autoregressive framework that transforms multi-shot video generation into an online directing process. CausalCine generates causally across shot changes, accepts dynamic prompts on the fly, and reuses context without regenerating previous shots. To achieve this, we first train a causal base model on native multi-shot sequences to learn complex shot transitions prior to acceleration. We then propose Content-Aware Memory Routing (CAMR), which dynamically retrieves historical KV entries according to attention-based relevance scores rather than temporal proximity, preserving cross-shot coherence under bounded active memory. Finally, we distill the causal base model into a few-step generator for real-time interactive generation. Extensive experiments demonstrate that CausalCine significantly outperforms autoregressive baselines and approaches the capability of bidirectional models while unlocking the streaming interactivity of causal generation.

For agents to learn continuously from interaction with the world at test time, they must be able to explore effectively, acquire new world knowledge and skills, retain relevant episodic experiences, and plan over long horizons. To evaluate these key abilities of test-time continual learning agents, we introduce AgentOdyssey, a novel evaluation framework that procedurally generates open-ended text games with rich entities, world dynamics, and long-horizon tasks. Critically, AgentOdyssey goes beyond the conventional machine learning assumption that learning does not occur at test time by placing agents in a continuous, long-horizon setting that interleaves learning and inference throughout deployment. We further propose a multifaceted evaluation methodology that measures not only game progress but also offers diagnostic tests on world knowledge acquisition, episodic memory, object and action exploration, action diversity, and model cost. We evaluate diverse agent paradigms in the generated games. Our experimental results reveal critical limits in agents' key abilities, as well as factors that influence their meaningful horizon. Although performance scales with stronger base models, even the top agent remains far below human performance, leaving substantial headroom for improvement. Among agent mechanisms, we find that short-term memory benefits multiple agent paradigms and is an important component of agent test-time training.

Generative modeling of longitudinal Electronic Health Records is increasingly important for privacy-preserving research, yet standard autoregressive models tend to underrepresent the co-occurrence structure of tail events (i.e., diseases, symptoms), reducing the fidelity and faithfulness of generated data for rare subpopulations. To this end, we propose ADAPCLA framework, which enables generative models to adaptively fit and generate EHR data through a data distribution-aware training strategy; this is achieved by internalizing data knowledge parameters by simulated annealing training. It also supports training-free adaptation to a diverse clinical population for generation through zero-shot distribution control. Moreover, our theoretical analysis characterizes rare-code logit updates through the label-wise empirical NTK and derives a prior-internalization bound for how annealing speed and NTK conditioning affect retained prior signals. Experiments on real-world data show that ADAPCLA achieves consistent gains in tail plausibility, downstream utility, and zero-shot control; in particular, it improves TailPairSeen over HALO by 114.2\% on MIMIC-III and 65.1\% on MIMIC-IV, outperforms GPT-style generation by 3.5\% F1 for zero-shot cross-population adaptation.