NeurIPS 2026 论文列表

Unsupervised video object-centric learning requires temporally consistent representations for downstream tasks. Existing methods mainly build temporal consistency on the assumption that adjacent frames share the same object composition. However, this local assumption is insufficient for real videos. Some objects remain semantically consistent over much longer temporal spans, while object composition may also change as objects appear or disappear. Fixed-slot recurrent Slot Attention architectures struggle to accommodate such visibility changes, limiting their ability to learn stronger temporal consistency beyond adjacent frames. We propose Dynamic Prototypical Contrast (DynaProto) to address these challenges. DynaProto introduces a dynamic recurrent Slot Attention architecture that determines whether each slot should be activated at each frame, thereby adapting to changing object visibility over longer temporal scopes. In addition, DynaProto introduces a prototypical contrastive objective that provides stronger clip-level temporal consistency supervision. Together, these two components enable DynaProto to learn more stable slot binding over extended temporal ranges. Extensive experiments show that DynaProto learns more temporally consistent slot representations and substantially outperforms prior methods on object discovery and object property prediction on real-world datasets, including an 11.0\% mBO improvement on YouTube-VIS.

Long-horizon forecasting of dynamical time series remains a fundamental challenge in machine learning, as accurate predictions require learning the underlying dynamics. While recent time-series foundation models leveraging in-context learning (ICL) have shown promise, we show that they fail to genuinely capture the governing dynamics. By rewriting the context of a time series as a collection of lagged input-output pairs, we frame dynamical time-series forecasting as a posterior inference problem over the space of structural causal models (SCMs), arguing that the underlying dynamic structure is a natural inductive bias. We propose DynaPFN, which uses pretrained tabular Prior-Fitted Networks (PFNs) to approximate this posterior through ICL, augmented with a library of dynamical features that permit simpler representations of complex systems. Even though the pretrained models have never seen time series data, DynaPFN outperforms existing zero-shot forecasting baselines on established benchmarks of chaotic time series and real-world datasets, demonstrating that incorporating structural priors over system dynamics is a valid approach for robust long-horizon forecasting.

Imaging of dynamic cellular responses to perturbations requires live-cell measurements, posing a challenge for fluorescence-based assays that can visualize molecular phenotypes by specific labeling but often affect cell health. Virtual staining helps by predicting fluorescence-like signals from label-free microscopy. Although many models have been developed for this task, their performance on volumetric time-lapse data remains poorly characterized. We address this gap by introducing DynaCell, an evaluation framework for 3D time-lapse virtual staining of live cells, comprising a new paired label-free and fluorescence 3D time-lapse imaging dataset, a suite of baseline models, and a three-tier metric panel measuring pixel-level fidelity, organelle segmentation performance, and single-cell phenotypic similarity. Using DynaCell, we study model behavior across 2 cell types and microscopes, 4 organelles, and 3 perturbation states. We find that regression baselines better predict the spatial localization of target organelles and are more robust to label-free input distribution shifts, while the generative baseline better captures population-level phenotype distribution across cells and organelles. These differences expose trade-offs between preservation of structural and phenotypic fidelity that are obscured by single-metric evaluation, and suggest how virtual staining can support downstream tasks such as localization, segmentation, tracking, and phenotypic profiling. DynaCell provides reusable data, code, checkpoints, and documentation for assessing virtual stains for live-cell biological measurements.

Safety is a key problem in reinforcement learning, especially during exploration. While safety filters are promising to address this issue, they are ill-suited for high-dimensional systems with unknown dynamics. We propose Dyna-style Safety Augmented Reinforcement Learning (Dyna-SAuR), a novel algorithm that learns a control policy as well as a scalable safety filter requiring minimal domain knowledge by leveraging a learned uncertainty-aware dynamics model. A novel filter learning formulation enforces to avoid failures and areas of high model uncertainty. As joint deployment of the control policy and filter allow to safely collect data and improve the dynamics model, certain areas grow and filter conservatism decreases over time. Dyna-SAuR reduces failures compared to state-of-the-art methods by two orders of magnitude on goal-reaching CartPole and MuJoCo Walker.

Knowledge editing aims to update outdated or incorrect knowledge in large language models without full retraining. Existing parameter-editing methods remain limited for long-sequence knowledge: single-point editors provide insufficient coverage for extended targets, while fixed chunk-based editors distribute editing effort uniformly across rigid predefined windows, failing to prioritize difficult, information-dense tokens. In this work, we show that token entropy is closely associated with both editing difficulty and semantic importance: high-entropy regions are harder to edit and often contain key factual content. Motivated by this observation, we propose DynEdit, a dynamic framework for long-sequence knowledge editing. DynEdit uses token-level entropy to guide non-uniform sliding, increasing the coverage of difficult high-entropy regions, and further applies entropy-guided token-level optimization to strengthen updates on information-dense tokens within each editing window. Experiments on three benchmarks and two backbone models show that DynEdit consistently outperforms competitive baselines across diverse long-form editing settings, achieving gains of up to +14.0 BLEU points on challenging paraphrase queries and near-saturated semantic scores on several diverse knowledge domains.

While novel view synthesis (NVS) for dynamic scenes has seen significant progress, reconstructing temporally consistent geometric surfaces remains a challenge. Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) offer powerful dynamic scene rendering capabilities; however, relying solely on photometric optimization often leads to geometric ambiguities. This results in discontinuous surfaces, severe artifacts, and broken surfaces over time. To address these limitations, we present DySurface, a novel framework that bridges the effectiveness of explicit Gaussians with the geometric fidelity of implicit Signed Distance Functions (SDFs) in dynamic scenes. Our approach tackles the structural discrepancy between the forward deformation of 3DGS ($canonical \rightarrow dynamic$) and the backward deformation required for volumetric SDF rendering ($dynamic \rightarrow canonical$). Specifically, we propose the VoxGS-DSDF branch that leverages deformed Gaussians to construct a dynamic sparse voxel grid, providing explicit geometric guidance to the implicit SDF field. This explicit anchoring effectively regularizes the volumetric rendering process, significantly improving surface reconstruction quality, containing watertight boundaries and detailed representations. Quantitative and qualitative experiments demonstrate that DySurface significantly outperforms state-of-the-art baselines in geometric accuracy metrics while maintaining competitive rendering performance. Codes will be publicly available.

Large-scale foundation models achieve strong performance across diverse tasks, but their size makes inference costly, largely due to dense matrix multiplications. Prior work reduces this cost by replacing dense weight matrices with efficient structured forms such as low-rank factorizations. However, these methods approximate weights rather than the output activations that determine inference accuracy. Consequently, small weight-space errors can be amplified by input activations, producing large output errors. In this work, we propose DyRA, an input-adaptive method that improves structured matrix multiplication approximation by correcting residual output errors during inference. We show that matrix multiplication can be approximated more effectively by directly optimizing low-rank factors of the output. DyRA builds on this insight by dynamically approximating and correcting the output error introduced by structured weight approximations. This combines efficient structured computation with input-dependent correction, yielding a more faithful approximation of full matrix multiplication under the same computational budget. Across vision, speech, and language models, DyRA consistently improves the accuracy–efficiency trade-off over structured weight approximations alone. Notably, DyRA achieves a $1.5\times$ end-to-end GPU speedup for DINOv3 while reducing accuracy degradation by more than $3\times$ relative to weight-only baselines.

Consistency-based generative models like Shortcut and MeanFlow achieve impressive results via a target-aware design for solving the Probability Flow ODE (PF-ODE). Typically, such methods introduce a target time $r$ alongside the current time $t$ to modulate outputs between a local multi-step derivative ($r = t$) and a global few-step integral ($r = 0$). However, the conventional "one input, one output" paradigm enforces a partition of the training budget, often allocating a significant portion (e.g., 75% in MeanFlow) solely to the multi-step objective for stability. This separation forces a trade-off: allocating sufficient samples to the multi-step objective leaves the few-step generation undertrained, which harms convergence and limits scalability.To this end, we propose Duality Models (DuMo) via a "one input, dual output" paradigm. Using a shared backbone with dual heads, DuMo simultaneously predicts velocity $\mathbf{v}_t$ and flow-map $\mathbf{u}_t$ from a single input $\mathbf{x}_t$. This applies geometric constraints from the multi-step objective to every sample, bounding the few-step estimation without separating training objectives, thereby significantly improving stability and efficiency. On ImageNet $256 \times 256$, a 679M Diffusion Transformer with SD-VAE achieves a state-of-the-art (SOTA) FID of 1.79 in just 2 steps.Code will be publicly available.

Recent multimodal agents have shown promising progress in embodied planning and web interaction. However, existing evaluations often rely on simplified tasks with limited combinatorial structure, leaving it unclear whether current agents can robustly integrate perception and reasoning for multi-goal planning under temporal, physical, and multimodal constraints. We introduce DualWorldBench, a unified benchmark for evaluating agent planning under controlled symbolic--temporal and grounded--topological complexity. Built around a shared pickup--delivery interface, DualWorldBench consists of three complementary components: TextWorld, which isolates symbolic planning under varying task compositions and release dynamics; SimWorld, which evaluates grounded path planning under controlled topological variation with text, image, or multimodal inputs; and DualWorld-VQA, a diagnostic suite spanning perception, scene understanding, and planning. Experiments show that current state-of-the-art agents remain brittle under increasing combinatorial complexity. Multimodal input does not consistently improve planning and can even degrade performance, revealing difficulties in aligning symbolic task specifications with grounded topology. Diagnostics further show that models handle perception or understanding in isolation, but fail when decision-making requires integrating both. These findings establish DualWorldBench as a challenging testbed for agents with tighter perception--reasoning--planning integration. All datasets and code are publicly available at our project website https://dualworld.netlify.app/.

Existing graph neural network-based methods for Boolean satisfiability (SAT) solving are generally constrained by limited local receptive fields, making it difficult to adequately capture the long-range dependencies between variables and clauses in SAT instances. Moreover, these methods overlook the multi-solution nature of SAT instances. Consequently, their expressive power and generalization performance remain limited. To address these issues, we propose DualSAT, a novel neural framework for the SAT problem, designed to enhance the phase selection heuristic of SAT solvers. Specifically, DualSAT adopts a novel dual-branch graph neural network that separately encodes local structural information and global structural information in the graph, and then fuses them through an attention-based cross-branch feature fusion module. An assignment decision head is further introduced to generate the initial assignment prediction of variables. This prediction process requires only a single forward pass, achieving a favorable balance between expressive power and inference efficiency. Furthermore, to account for the multiple solutions nature of SAT instances, we design a two-stage training strategy that combines supervised and unsupervised learning to improve the generalization performance of the model. Experimental results on multiple datasets show that, as an end-to-end assignment prediction model, DualSAT significantly outperforms NeuroSAT and NeuroBack. In addition, we integrate DualSAT into the classical SAT solver CaDiCaL, where it increases the number of solved instances by up to 11.6% and reduces the number of conflicts by up to 9.21% on SAT competition problem sets.

Modern RL post-training methods such as GRPO and DAPO train on $N$ response sequences of $R$ tokens sampled from a shared prompt of $P$ tokens, but standard FlashAttention replicates all $P$ prompt tokens $N$ times across both forward and backward passes --- duplicating compute and memory on identical hidden states. In large-rollout, long-context RL training ($N{\geq}16$, $P{\geq}8\text{K}$), this redundancy dominates the policy update cost. We observe that in decoder-only models, causal masking makes prompt representations invariant across sequences at every layer, so all per-token operations (norms, projections, MLP) and attention can process the prompt once --- a property not yet exploited at the kernel level for training. We propose \textbf{DualKV}, the first FlashAttention kernel variant that eliminates shared-prompt replication during RL training, via (1) fused CUDA forward and backward kernels that iterate over two disjoint KV regions --- shared context and per-sequence response --- in a single kernel launch, and (2)~a data-pipeline redesign in veRL that repacks $N(P{+}R)$ tokens into $P{+}NR$ tokens per micro-batch, extending the token reduction from attention to the entire model by a factor $\rho = N(P{+}R)/(P{+}NR)$. DualKV is mathematically equivalent to standard attention and introduces no approximation. On Qwen3-8B GRPO training with 8$\times$H100 GPUs ($N{=}32$, 8K-context), DualKV achieves $1.63$--$2.10\times$ policy-update speedup, enables $2\times$ larger micro-batches, and raises MFU from $36\%$ to $76\%$. Similar gains hold for DAPO ($2.47\times$ speedup, $77\%$ MFU). At 30B MoE scale on 16$\times$H100, DualKV achieves $3.82\times$ policy-update and $3.38\times$ end-to-end step speedup over FlashAttention (which requires 4-way Ulysses sequence parallelism to avoid OOM).

Drifting Models have recently achieved state-of-the-art performance in one-step image generation by training generators to follow corrective vector fields. However, their standard reverse-style drift improves samples from the model side, which can severely underutilize training samples and bottleneck early convergence. To address this, we introduce DualDrift, a unified framework that augments the original reverse drift with a novel forward drift. The forward drift lets each data sample assign correction to generated rollouts, providing denser data-side signals and accelerating early training. At the same time, this dense data-side signal can overemphasize dominant regions when used alone. DualDrift therefore uses a parameter-free scheduler driven by empirical batch statistics to balance the fast early correction of forward drift with the stable refinement of reverse drift. Evaluated on ImageNet $256\times256$, DualDrift achieves faster convergence and significantly improves over the baseline Drifting Models in both FID and Inception Score.

This paper develops a dual-space preconditioning framework for variational inequalities and root-finding problems beyond classical cocoercivity and Lipschitz continuity assumptions. To this end, we introduce both a novel cocoercivity-type condition and a relaxed Lipschitz continuity condition, and derive convergence of deterministic and stochastic methods under these generalized conditions. For isotropic preconditioners, the resulting algorithms can be interpreted as variable-stepsize variants of classical methods, recovering and extending recent analyses based on $(L_0, L_1)$-type growth conditions while allowing simpler proofs and larger stepsize ranges. We also show that the framework naturally handles constrained variational inequalities through weak Minty-type arguments. Finally, we study the associated continuous-time dynamics and derive a variational and optimal-control interpretation of the preconditioned flow. Overall, the results show that dual-space preconditioning provides a flexible mechanism for designing and analyzing operator methods beyond the standard Euclidean Lipschitz setting.

Diffusion models achieve state-of-the-art generative performance but suffer from high computational costs during inference due to the repeated evaluation of a heavy neural network. In this work, we propose Dual-Rate Diffusion, a method to accelerate sampling by interleaving the execution of a heavy high-capacity context encoder and a light efficient denoising model. The context encoder is evaluated sparsely to extract high-dimensional features, which are effectively reused by the light denoising model at every step to refine the sample efficiently. This approach significantly accelerates inference without compromising sample quality. On ImageNet benchmarks, Dual-Rate Diffusion matches the performance of standard baselines while reducing computational cost by a factor of $2$--$4$. Furthermore, we demonstrate that our method is compatible with distillation techniques, such as Moment Matching Distillation, enabling further efficiency gains in few-step generation.

Adapting Multimodal Large Language Models (MLLMs) to specialized and diverse downstream applications necessitates fine-tuning, typically through parameter-efficient methods such as *Low-Rank Adaptation* (LoRA). However, fine-tuning MLLMs severely degrades pre-trained upstream capabilities, a phenomenon known as *catastrophic forgetting*. Existing Methods primarily rely on parameter pruning to strike a balance between the downstream adaptation and upstream knowledge retention. Empirical analysis indicates that these methods are subject to *parameter space entanglement*, wherein the imposition of downstream adaptation and upstream retention within a unified parameter space results in suboptimal outcomes for both objectives. To address this, we introduce Dual-Pronged LoRA, a novel fine-tuning framework for MLLMs featuring two components. (1) *Task-Decoupled Contextual Routing* (TDCR) makes the first attempt to explicitly decouple tasks by modeling a downstream contextual support region. It adaptively activates the LoRA branch exclusively for downstream-relevant inputs and bypasses it otherwise, thereby preventing downstream adaptation from interfering with pre-trained knowledge. (2) *Downstream-Performance-Oriented Rank Sculpting* (DPO-RS) maximizes downstream adaptation performance by dynamically sculpting ranks via gradient energy. It effectively strengthens critical layers while compressing redundant ones and is theoretically grounded in the minimization of information loss. Extensive experimental results indicate that Dual-Pronged LoRA achieves near-zero forgetting while attaining new state-of-the-art downstream performance.

For decades, philosophers and musicologists have debated which features of a performance are constitutive of the work and which are expressions of how it is being played. Computational evidence has been hard to assemble: audio embedding models conflate work identity with performance style, and existing interpretability tools for music generators treat learned representations as flat dictionaries that mix the two. We turn the question into an empirical one by introducing the dual-contrastive sparse autoencoder (DC-SAE), a two-branch sparse autoencoder that uses cheap work-level metadata to factor a generative music transformer's residual stream into work-identity and performance-content subspaces. Across jazz-standard, classical-work, and pop-cover corpora, the resulting decomposition is supported by probes and feature galleries that surface musically interpretable concepts on each side. Without performer supervision, the variation branch acquires structure related to performer identity, and steering along these directions can shift the perceived performer of generated audio while preserving the underlying work. Together, these results show that a generative music model has internalized a representation of interpretation that can be recovered, named, and steered.

Decoder-only Transformers compute attention over the KV cache of preceding tokens. Keys (and Values) are typically represented with the same dimensionality, regardless of its distance from the prediction target. In natural language, however, the next word is most strongly influenced by the immediately preceding tokens. We hypothesize that local and distant tokens impose asymmetric demands on representational capacity: local tokens are more critical for predicting immediate outputs and thus require richer representations, whereas distant tokens primarily serve as long-range memory, for which lower-dimensional representations may suffice. We formalize this idea as Distance-Adaptive Representation (DAR), implemented in a controlled setting that preserves full-dimensional representations within a local context window while assigning reduced-dimensional representations (e.g. 1/4 of the original dimensionality) to tokens beyond that window. Across multiple pretraining scales (70M to 410M parameters), as well as continued supervised fine-tuning on a 1B-scale model, this approach closely matches the performance of full-dimensional baselines. In contrast, uniformly reducing dimensionality across all token positions leads to worse performance. These results challenge the common assumption that key and value dimensionality should be uniform across token positions. Our findings suggest a new direction for designing attention architectures that adaptively allocate representational capacity across the sequence, enabling further reductions in KV cache during inference.

Building state-of-the-art (SOTA) predictive models for drug discovery requires expensive search over tools, architectures, and training strategies. Current LLM-based agents can find SOTA solutions through extensive trial and error, but they do not retain the experience accumulated along the way and therefore pay the full search cost on every new task. We propose DrugSAGE (Self-evolving Agent Experience), a framework that accumulates and reuses experience across tasks to build SOTA drug discovery models efficiently. DrugSAGE maintains a cross-task memory of verified skills, statistical evidence about effective strategies, and a record of recurring errors and their fixes. In some cases, DrugSAGE transfers a working solution directly without test-time search. In 33 molecular property prediction tasks, DrugSAGE ranks first among nine SOTA agents in a single-task setting. With memory accumulated from 16 smaller tasks, DrugSAGE achieves a averaged normalized score of 0.935 on 17 held-out tasks in a cross-task evaluation setting and outperforms all baseline agents by 10-30\% in a zero-test-time search regime. In summary, our work shows the advantage of cross-task memory for efficient SOTA model development in drug discovery.

Key-value (KV) cache management has become a critical bottleneck for scaling Large Language Models (LLMs) to long-context scenarios due to linear memory growth. While recent eviction methods have moved beyond simple attention-score heuristics to incorporate value representations, they typically face a fundamental trade-off: the underlying optimization for minimizing output perturbation is an NP-hard combinatorial problem, leading to mathematically suboptimal heuristics or implementations that are difficult to kernelize. In this work, we propose DropKV, a simple principled eviction framework based on Decoupled Residual-Output Perturbation. By decoupling the joint eviction decision into independent per-token scoring, DropKV circumvents combinatorial intractability and admits a provable constant-factor approximation guarantee under a cone condition that we empirically verify on three long-context LLMs. To ensure practical viability, we provide a high-performance implementation using fused Triton kernels that avoid materializing the full attention matrix, delivering a 19.8$\times$ scoring-kernel speedup that translates to up to a 9.9\% matched-batch end-to-end prefill speedup. Extensive evaluations on the RULER and LongBench benchmarks demonstrate that DropKV consistently outperforms state-of-the-art baselines at equivalent cache budgets, achieving the lowest inference latency among eviction methods while also remaining faster than the dynamic cache baseline.