NeurIPS 2026 论文列表

While representation alignment with self-supervised models has been shown to improve diffusion model training, its potential for enhancing inference-time conditioning remains largely unexplored. We introduce Representation-Aligned Guidance (REPA-G), a framework that leverages these aligned representations, with rich semantic properties, to enable test-time conditioning from features in generation. By optimizing a similarity objective (the *potential*) at inference, we steer the denoising process toward a conditioned representation extracted from a pre-trained feature extractor. Our method provides versatile control at multiple levels of granularity, ranging from patch level matching via single patches to broad semantic guidance using global image feature tokens. We further extend this to multi-concept composition, allowing for the faithful combination of distinct concepts. REPA-G operates entirely at inference time with no additional training required, offering a flexible and precise alternative to often ambiguous text prompts or coarse class labels. Our approach achieves high-quality, diverse generations on ImageNet and COCO.

Test-time compute has emerged as a promising strategy to enhance the reasoning abilities of large language models (LLMs). However, this strategy has in turn increased how much users pay cloud-based providers offering LLM-as-a-service, since providers charge users for the amount of test-time compute they use to generate an output. In our work, we show that the market of LLM-as-a-service is socially inefficient–providers have a financial incentive to increase the amount of test-time compute, even if this increase contributes little to the quality of the outputs. To address this inefficiency, we introduce a reverse second-price auction mechanism where providers bid their offered price and (expected) quality for the opportunity to serve a user, and users pay proportionally to the marginal value generated by the winning provider relative to the second-highest bidder. To illustrate and complement our theoretical results, we conduct experiments with multiple instruct models from the $\texttt{Llama}$ and $\texttt{Qwen}$ families, as well as reasoning models distilled from $\texttt{DeepSeek-R1}$, on benchmark datasets covering mathematics, science, and open-ended question answering.

Pre-trained vision-language models (VLMs) enable zero-shot classification by matching images with text prompts, yet their raw confidence scores are unreliable for out-of-distribution (OOD) detection. We identify a confidence bottleneck: zero-shot VLMs produce posteriors whose maximum confidence yields much weaker ID/OOD separation than supervised classifiers trained on the target label space. To theoretically explain this gap, we relate VLM confidence to a well-separated supervised reference through a Markov-kernel transformation and analyze single-threshold detection using the Youden/KS index. Our analysis shows why raw VLM confidence is unlikely to recover the clean separation achieved by supervised classifiers. Motivated by the observation that class-wise posterior distributions contain useful structure beyond maximum confidence, we propose Self-Reinforced Optimal Transport (SROT), a label-free test-time framework that reshapes VLM posteriors with optimal transport and adaptively trains a lightweight OOD detector from rectified pseudo-labels. The detector further feeds open-set evidence back into posterior alignment, forming a self-reinforcing loop between distribution-level rectification and feature-level OOD learning. Extensive experiments on standard zero-shot OOD detection benchmarks show that SROT achieves state-of-the-art performance, with ablations validating posterior alignment, adaptive detector training, and self-reinforcement.

Test-time training (TTT) enhances model performance by explicitly updating designated parameters prior to each prediction to adapt to the test data. While TTT has demonstrated considerable empirical success, its theoretical underpinnings remain limited, particularly for nonlinear models. In this paper, we investigate the combination of TTT with in-context learning (ICL), where the model is given a few examples from the target distribution at inference time. We analyze this framework in the setting of single-index models, where the feature vector is drawn from a hidden low-dimensional subspace. For single-layer transformers trained with gradient-based algorithms and adopting TTT, we establish an upper bound on the prediction risk. Our theory reveals that TTT enables the single-layer transformers to adapt to both the feature vector and the link function, which vary across tasks. This creates a sharp contrast with ICL alone, which is theoretically difficult to adapt to shifts in the link function. Moreover, we provide the convergence rate with respect to the data length, showing the predictive error can be driven arbitrarily close to the noise level as the context size and the network width grow.

Language models are increasingly deployed alongside test-time search. We argue that we should therefore shift the role of RL post-training from converging on a single best response to producing a diverse pool of competent candidates. Standard policy gradient methods such as GRPO optimize a fixed scalar reward and drive the policy toward near-duplicate responses, erasing the diversity search needs. We propose Vector Policy Optimization (VPO), which exploits the fact that practical rewards are vector-valued, e.g., per-test-case correctness, per-criterion ratings, or per-hop credit. Instead of collapsing these into one scalarization, VPO samples randomized scalarizations from a distribution over the reward simplex, incentivizing candidates to specialize to different trade-offs along the Pareto front. The model emits multiple candidates per prompt, with each conditioning on the previous ones, allowing for directed diversification. VPO is a drop-in replacement for the GRPO advantage estimator. Across four domains, VPO consistently improves test-time best@$k$ over scalar baselines, with gains widening as the test-time budget grows.

We study Flow Matching in a semi-discrete setting where a Gaussian source is transported toward a discrete target supported on finitely many points. This semi-discrete regime is the theoretical setting behind the use of Flow Matching for generative modeling, where the target distribution is represented by a finite dataset. In this semi-discrete regime, the exact Flow Matching velocity field is available in closed form, which makes it possible to analyze the geometry induced by the terminal flow map independently of optimization and approximation effects. We investigate the terminal assignment regions, namely the preimages of the target atoms under the terminal flow. We show that these regions are open, simply connected and, under an additional assumption, homeomorphic to the unit ball. At the same time, a planar four-point example shows that these cells can differ sharply from Laguerre cells arising in semi-discrete optimal transport: they may be non-convex, have curved boundaries, and exhibit different boundedness and adjacency patterns. These results clarify the geometry intrinsically induced by the exact semi-discrete Flow Matching objective before neural approximation enters the picture.

Recent text-to-image models have made substantial progress in realism, aesthetics, and prompt-image alignment. Yet visually appealing images can still violate real-world plausibility, exhibiting malformed object structures, impossible anatomy, physically implausible interactions, or impossible spatial relationships. These failures are not well captured by existing quality, aesthetics, preference, or alignment metrics. To address this gap, we introduce __TerraVis__, a framework for evaluating __world-grounded visual consistency__ in generated images. TerraVis defines a structured taxonomy of world-consistency violations covering object-, interaction-, and scene-level failures, and evaluates images through a multi-stage workflow. Specifically, TerraVis uses an MLLM to first perform eligibility checking, determining whether an image is suitable for world-consistency evaluation. It then conducts taxonomy-guided violation detection over fine-grained violation types and classifies detected violations as minor or severe to derive an overall world-consistency score. Across diverse open-source and proprietary text-to-image models on two widely used benchmarks, TerraVis shows the strongest correlation with human judgments of world consistency among the evaluated metrics. Our results further reveal that models with high quality, aesthetics, preference or alignment scores can still exhibit frequent world-consistency failures. TerraVis therefore provides a complementary evaluation perspective, enabling fine-grained diagnosis of generated images and model comparison beyond existing evaluation dimensions.

Open‑vocabulary segmentation (OVS) enables models to segment a variety of concepts specified in natural language, removing the need for fixed label sets and fine-tuning. While recent OVS approaches have shown strong performance on natural images, they transfer poorly to Earth observation (EO) data due to differing semantics, scales, sensing modality, and geographic context. To date, large‑scale EO‑specific datasets and benchmarks for OVS remain scarce. We, therefore, introduce TerraMesh-Masks, a large-scale open-vocabulary segmentation dataset for EO, comprising 16 million binary masks aligned with nine million multimodal TerraMesh samples and annotated using OpenStreetMap and Overture-derived semantics. In addition, we provide an expert-reviewed evaluation set of 523 samples covering over 200 classes to enable robust evaluation. To demonstrate and analyze the impact of TerraMesh-Masks, we build an open-vocabulary version of TerraMind. Our results demonstrate significantly improved performance over general-purpose state-of-the-art models reaching 35.15% IoU versus 16.29% for DINOv3.txt and 11.33% for SAM-3. We release the dataset publicly under a permissive license at https://huggingface.co/datasets/ibm-esa-geospatial/TerraMesh-Masks.

We introduce TerminalWorld, a scalable data engine that automatically reverse-engineers high-fidelity evaluation tasks from "in-the-wild" terminal recordings. Processing 80,870 terminal recordings, the engine yields a full benchmark of 1,530 validated tasks, spanning 19 real-world categories, ranging from short everyday operations to workflows exceeding 50 steps, and covering 1,280 unique commands. From these, we curate a Verified subset of 200 representative, manually reviewed tasks. Comprehensive benchmarking on TerminalWorld-Verified across eight frontier models and six agents reveals that current systems still struggle with authentic terminal workflows, achieving a maximum pass rate of only 62.5%. Moreover, TerminalWorld captures real-world terminal capabilities distinct from existing expert-curated benchmarks (e.g., Terminal-Bench), with only a weak correlation to their scores (Pearson $r=0.20$). The automated engine makes TerminalWorld authentic and scalable by construction, enabling it to evaluate agents in real-world terminal environments as developer practices evolve. Data and code are available at https://github.com/EuniAI/TerminalWorld.

Most large language models (LLMs) run on external clouds: users send a prompt, pay for inference, and must trust that the remote GPU executes the LLM without any adversarial tampering. We critically ask how to achieve verifiable LLM inference, where a prover (the service) must convince a verifier (the client) that an inference was run correctly without rerunning the LLM. Existing cryptographic works are too slow at the LLM scale, while non-cryptographic ones require a strong verifier GPU. We propose TensorCommitments (TCs), a tensor-native proof-of-inference scheme. TC binds the LLM inference to a commitment, an irreversible tag that breaks under tampering, organized in our multivariate Terkle Trees. For LLaMA2, TC adds only 0.97% prover and 0.12% verifier time over inference while improving robustness to tailored attacks by up to 48% over the best prior work requiring a verifier GPU.

Transformers for temporal sequences require mapping sampled signals into sequences of vectors, called tokens. This is done by partitioning the sequence into local temporal windows (patches). We show that the tokenization step induces an approximation--estimation tradeoff. Smaller windows yield higher-variance statistical estimates because each token is supported by fewer samples, while larger windows stabilize estimation but increase the amount of information that must be summarized. Motivated by this tradeoff, we propose a lightweight tokenizer that constructs multiple estimates for each token and sums them, without changing the Transformer backbone. Across 12 datasets spanning four modalities, our method improves performance and reduces sensitivity to temporal scale across tasks.

Neurobiological circuits may be energy efficient in part because there is a division of labor: different subsystems compute or represent different things. In light of this, it is interesting that a major difference between subsystems is the time scale on which they typically vary. Here, we argue that temporal smoothness constraints---the idea that it may be energetically costly for systems to vary much more quickly or slowly than their typical time scale---imply a certain division of labor with respect to temporal stimulus features. In particular, slow subsystems ought to represent slowly varying stimulus features, and fast subsystems ought to represent quickly varying features. We formulate a novel efficient coding model that we use to investigate this claim, and exactly solve for the optimal division of labor. In addition to finding that slower subsystems ought to represent more slowly varying temporal features, we find surprising structure of mathematical interest: optimal codes have subsystems which represent orthogonal temporal features, and which correspond to the eigenfunctions of a Sturm-Liouville problem.

Recent advances in generative video models have significantly improved visual realism, making detection increasingly dependent on subtle temporal inconsistencies arising from imperfect frame-to-frame coherence. However, these signals are weak and spatially localized, while most of the video content is redundant. This creates a fundamental mismatch: existing detectors rely on heavy backbones to model full videos, assuming increased capacity can capture such cues, which leads to high computational cost and limited scalability. In this work, we rethink AI-generated video detection from a representation perspective. Instead of modeling entire videos, we propose to explicitly restructure them into compact temporal slice representations that isolate informative temporal evolution. Based on this idea, we introduce a lightweight framework using frequency-aware temporal slice consistency learning, where informative spatial rows and columns are aggregated over time to form slice-domain inputs. This reparameterization suppresses redundant appearance while preserving discriminative temporal dynamics, enabling efficient and targeted detection. By constructing slices from both high- and low-frequency regions, our method exposes complementary temporal artifacts, including unstable details and abnormal motion smoothness. We instantiate this idea in a $\textit{Temporal Slice Consistency Network}$, which integrates task-driven slice localization, hierarchical positional encoding, and a lightweight Transformer to model cross-slice and cross-frequency dependencies with minimal overhead. We further introduce $\textit{AI-Artist}$, a new benchmark with artist-curated videos from recent high-fidelity generators, including Seedance and Kling. Experiments on GenVideo and AI-Artist show that our method achieves strong performance while requiring over 400× fewer FLOPs and 3.6× faster inference than the strong video-based baseline ReStraV, enabling scalable and practical deployment.

Flow matching has shown strong potential in generative tasks, while its optimization via Reinforcement Learning (RL) remains underexplored, especially in continuous-discrete mixed settings. Moreover, flow matching is formulated as an Ordinary Differential Equation (ODE), whose deterministic trajectories do not naturally support RL optimization. Introducing stochasticity by converting the entire trajectory into a Stochastic Differential Equation (SDE) is a common practice. Nevertheless, not all timesteps contribute to the final outcomes, and redundant exploration introduces additional noise. Furthermore, in multi-objective settings, some objectives may dominate the optimization while others receive insufficient updates. In this study, we propose an RL-guided continuous-discrete flow matching framework with temporal selective exploration for 3D de novo molecular design. Specifically, our method jointly optimizes continuous and discrete flow matching for atomic coordinates and molecular identities, respectively, while restricting exploration to timesteps that primarily contribute to target properties, reducing ineffective exploration. We also introduce a reward-based advantage calibration mechanism to mitigate objective dominance in multi-objective optimization. The framework is applied to design inhibitors for both the well-studied target protein EGFR and the challenging Pin1 protein. We identify some novel candidate inhibitors, supported by in silico validation.

Dataset distillation condenses large-scale training sets into compact synthetic data. Recent frozen-feature distillation methods, such as Linear Gradient Matching, make ImageNet-scale synthesis feasible by matching gradients in pretrained representation spaces, but their reliance on exhaustive spatial view ensembles incurs memory and I/O costs that grow linearly with the number of augmentations. We propose Temporal Prototype Alignment (TPA), an online temporal gradient estimation framework for frozen-feature dataset distillation. TPA replaces spatial Monte Carlo gradient averaging with a temporal prototype estimator that accumulates stable semantic signals across optimization steps. An adaptive momentum gate modulates the estimator according to local signal reliability, reducing gradient variance while keeping memory complexity independent of the number of spatial views. To improve cross-architecture transfer, we introduce a consensus alignment objective that distills shared structure from multiple pretrained teachers while mitigating model-specific artifacts. Across ImageNet-scale and fine-grained benchmarks, TPA substantially reduces memory overhead, enables distillation on consumer-grade GPUs, and improves transfer to heterogeneous CNN and Vision Transformer backbones. These results suggest that online temporal estimation provides an efficient alternative to spatial ensembling for frozen-feature dataset distillation.

Flow-matching-style generative models learn time-dependent vector fields, but standard objectives supervise each timestep independently, leaving shared-path temporal structure unused. We introduce *Temporal Pair Consistency* (TPC), a simple training-time objective that couples velocity predictions at paired timesteps along the same probability path. TPC leaves the architecture, probability path, sampler, and inference-time computation unchanged. Under matched backbones, training budgets, solvers, and numbers of function evaluations, TPC improves sample quality across flow matching, OT-CFM, rectified flow, DiT-style backbones, and MeanFlow. On CIFAR-10, TPC improves FM from 6.35 to 3.19 FID and OT-CFM from 3.58 to 2.90 FID at the same NFE. Mechanism diagnostics show lower measured gradient variance, higher paired-gradient correlation, and reduced temporal roughness, while random and local pairing controls are substantially weaker. Higher-resolution ImageNet experiments further show consistent gains for DiT-XL/2 at 256 and 512 resolution and for MeanFlow at 256 resolution, all at matched inference cost.

Deep learning models for electronic health records achieve strong predictive performance but provide little insight into *what clinical concepts* they represent internally. Post-hoc attribution methods produce per-cell importance scores that lack temporal coherence, cross-variable structure, and reusability across patients. We introduce the **Temporal Island Sparse Autoencoder** (TI-SAE), which learns a backbone-specific, patient-shared *clinical concept dictionary* from a frozen predictive model’s representations. Each dictionary entry (latent) is defined by learned temporal Gaussian islands—specifying *when* the concept is active—and a sparse variable-composition vector—specifying *which variables* are involved. A decoupled selector then identifies which dictionary entries drive each patient’s prediction, separating representation learning from attribution. We evaluate TI-SAE across five backbone architectures spanning different representation strategies on MIMIC-III and MIMIC-IV (three prediction tasks each). TI-SAE produces structured clinical concepts that go beyond transient attribution heatmaps while achieving strong performance on standard faithfulness benchmarks.

Distributed learning in embodied reinforcement-learning agents offers a degree of privacy by retaining raw sensor data on-device and transmitting only policy gradients to the server. Yet temporal structure can amplify this leakage beyond single-frame attacks. We introduce **T**emporal **R**econstruction **A**ttack on **C**onsecutive **E**ncodings (TRACE), an amortized temporal gradient-inversion attack that autoregressively reconstructs the sequence of private observation-action trajectories from per-step policy-learning gradients. The attack exploits two structural signals ignored by prior single-frame methods: (i) cross-time correlation between successive embodied gradients, which we formalize via a conditional mutual-information bound, and (ii) closed-form action recovery from policy-head gradient structure, which we prove exact when standard entropy regularization is sufficiently small. On held-out embodied scenes, TRACE reaches $18.8$ dB PSNR with near-perfect action recovery at $3$-$4.5$ ms per reconstructed frame, dominating the learning-based baseline across all reconstruction metrics and exceeding optimization attacks while running orders of magnitude faster. Defense experiments suggest that protecting temporal gradient streams may require sequence-aware privacy mechanisms.

Offline meta-reinforcement learning seeks to learn a policy that generalizes to new related tasks online. Context-based methods infer a task representation from transition histories, yet learning an effective task representation without supervision remains challenging. Existing methods relying on contrastive learning learn discriminative task representations, but fail to identify task-specific dynamics, while relying on reconstruction can be insufficient to model long-horizon dependencies, limiting generalization to new tasks. We investigate the impact of temporal consistency in latent space on task representation learning, showing that enforcing multi-step predictions in latent space encourages task representations that are able to capture task-dependent dynamics while preventing representation collapse. We provide theoretical analysis characterizing sources of error in value estimation and show through extensive experiments on MuJoCo, Contextual DeepMind Control, and MetaWorld benchmarks that temporal consistency significantly improves both zero-shot and few-shot generalization.