NeurIPS 2026 论文列表

LLM routing dispatches each user query to the most suitable model from a pool of candidates, balancing response quality against serving cost. As LLMs are increasingly deployed in conversational interfaces, routing decisions must now be made repeatedly within an ongoing session. Existing routers treat each query independently, implicitly assuming that the optimal model depends only on the current input. This assumption ignores two forms of session state that accumulate across turns: the dialogue state, which shapes query meaning and difficulty, and the \textit{serving state}, which determines each model's true \textit{serving cost} via KV-cache reuse. Neglecting dialogue state leads to misrouting when queries depend on prior turns, while ignoring serving state causes systematic cost overestimation that worsens as conversations grow longer. A practical multi-turn router must therefore capture dialogue state for accurate quality prediction, and track per-model serving state for accurate cost estimation. To address this gap, we introduce Chain-of-Route (CoR), the first state-aware routing framework for multi-turn conversations. CoR comprises two modules: a Dialogue State Propagator (DSP) that maintains a compact recurrent state to capture session-level context across turns, and a Serving State Tracker (SST) that tracks per-model KV-cache to compute true serving costs. Together with the framework, we construct a unified multi-turn routing benchmark from WildChat and LMSYS-Chat-1M with per-turn quality annotations across five LLMs. Experiments show that CoR achieves state-of-the-art quality-cost tradeoffs, reducing QNC by up to 20.3\% over the strongest baseline while improving response quality.

Dynamic Sparse Training (DST) is an effective paradigm for sparse to sparse training of neural network connectivity under a fixed parameter budget. Recent advances in network-science for AI introduced epitopological learning methods, such as Cannistraci-Hebb Training (CHT), demonstrating that network automata applied to the mere network topology enables gradient-free predictions of the sparse connectivity evolution that can trigger significant increase in task performance. However, these methods were currently developed for standard MLP-like fully-connected architectures and need significant rethinking to be extended to convolutional neural networks (CNNs) due to element-wise weight sharing and spatially repeated interactions. In this work, we bridge this gap by introducing epitopological learning DST for CNNs through channel-wise network modeling to address the prohibitive complexity of element-wise modeling, which treats each kernel-channel as a single network node. Results show orders-of-magnitude improvements in time complexity of link-regrowth when applied at channel-level with respect to element-level. We also enhance the channel-wise formulation with lightweight contextual modulation to improve expressivity. Extensive experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet with ResNet and VGG architectures demonstrate that our method achieves performance comparable to dense training using only 30% of the parameters, which significantly reduces computational cost. Furthermore, our approach improves robustness under noisy inputs. These results highlight the effectiveness of topology-aware sparse training and the benefit of decoupling graph construction from sparse update rules in convolutional networks.

Automatic neuron reconstruction from light microscopy images is a central problem in computational neuroanatomy. While recent methods have achieved encouraging results on local image blocks, it remains unclear whether such progress translates to reconstruction that is both structurally accurate and scalable to the whole-brain scale. We present CORAL, the first benchmark for structure-aware evaluation of automatic neuron reconstruction from light microscopy images at both local and whole-brain scales. Built on a high-quality whole-brain fMOST dataset with carefully curated annotations, CORAL establishes two progressive tasks: block-level reconstruction, which evaluates reconstruction methods under limited spatial context, and brain-wide reconstruction, which assesses complete neuron reconstruction at the whole-brain scale. To account for topological correctness beyond geometric distance similarity, we introduce a structure-aware metric based on fiber prediction. To further achieve complete neuron reconstruction across the entire brain, we develop a brain-wide neuron tracing framework that extends arbitrary local reconstruction methods to the whole-brain scale through an iterative local-to-global process. Using this benchmark, we provide the first structure-aware comparison of mainstream methods for local neuron reconstruction and further evaluate their performance in brain-wide reconstruction. Our results underscore the importance of structure-aware evaluation and the need for more robust methods for complete neuron reconstruction. All datasets and code are available at https://github.com/yang-ze-kang/CORAL.

Joint classification of hyperspectral imagery (HSI) and Light Detection and Ranging (LiDAR) data benefits from complementary spectral--geometric cues, but existing fusion methods largely assume locally consistent cross-modal features and symmetric pixel-wise trust, which is often violated in real urban scenes. We term this pixel-level phenomenon \emph{cross-modal local inconsistency} (CMI) and quantify it with a Strong-Edge CMI Ratio: dual-modal gradient analysis shows that over $16\%$ of strong-edge pixels exhibit single-modality dominance across three benchmarks, while ROI-level analysis reveals the class-entanglement induced by symmetric fusion. To address this issue, we propose \textbf{CMI-Trans} (\textbf{C}ross-\textbf{M}odal \textbf{I}nconsistency-aware \textbf{Trans}port), which promotes per-pixel reliability modeling from a post-hoc diagnostic to a structural signal for alignment and training. CMI-Trans combines Energy-Calibrated Reliability (ECR), Uncertainty-Shaped Transport Alignment (USTA) with Sinkhorn-regularized optimal transport, and Triadic Co-Optimization (TCO) that jointly optimizes classification, uncertainty consistency, and alignment stability on a Mamba-based fusion backbone. On Houston, Trento, and MUUFL, CMI-Trans achieves competitive overall accuracy and substantial gains on CMI-affected classes (e.g., $+17.90\%$ on Houston C10). The ECR-derived uncertainty score also separates reliable from erroneous predictions ($\rho(U,e)<0$; top-$20\%$ highest-$U$ error rate $\le 0.18\%$), while improving feature separability in shadow-induced CMI regions. These results show that explicitly modeling CMI through uncertainty-shaped adaptive alignment yields more accurate and interpretable multimodal remote sensing classification.

As sequence lengths continue to grow, distributed attention based on sequence parallel (e.g., Ring-Attention) has become a mainstream solution for large language models (LLMs). Nevertheless, inherent communication dependencies limit computation–communication overlap, reducing GPU utilization. Moreover, in cross-hardware (NUMA or node) architectures, due to imbalanced communication bandwidth, frequent communication synchronization will limit overall performance. Based on these insights, we propose CHARM+, Cross-Hardware Attention with Re-Merge Multistream Mechanism. We reorganize sequential communications into several levels and achieve intra-level parallelism and computation overlap. Furthermore, we decouple cross-hardware communication from intra-hardware communication, eliminating synchronization overhead caused by bandwidth imbalance. Forward and backward algorithms were evaluated on multiple hardware architectures and consistently outperformed state-of-the-art methods. Compared to Ring-Attention, our forward and backward algorithms achieve average speedups of $1.3\times$ and $1.7\times$ on a dual-NUMA PCIe architecture, and $2.8\times $ and $3.9\times$ on a dual-node NVLink architecture, respectively.

Despite growing interest in on-device LLM deployment, large-scale Mixture-of-Experts (MoE) models remain impractical on resource-constrained devices, as sparse expert activation still requires all expert weights to be memory-resident. To address this, we identify three key observations for on-device MoE inference: (a) cache hit rates are fundamentally bounded even under Belady’s optimal policy; (b) these bounds are insufficient for low-latency on-device inference; and (c) prior cache-aware routing based on binary cache presence can induce routing instability. Based on these insights, we propose CECAR (Cache & Expert Co-Aware Routing), a MoE inference system that jointly considers expert routing and cache management. Unlike prior cache-aware routing methods, CECAR uses an ML-based cache policy to predict the expected reuse distance of each expert and prioritizes routing and execution toward experts with smaller predicted reuse distances. On the Qwen3-30B-A3B model, CECAR achieves 12.5 tokens/s on a consumer-grade GPU with only 16 GB of VRAM, where non-resident experts are fetched from SSD, delivering a 4.5× speedup over conventional MoE inference and exceeding the decode speed of baselines that keep the entire model fully resident in DRAM.

Many studies argue that the success of sharpness-aware minimization (SAM) is due to the implicit regularization of the gradient norm or the eigenvalues of the Hessian. However, they cannot fully explain why the flatness-based measures do not always correlate well with generalization. For instance, we can easily construct a counterexample by only optimizing with hard examples. In this paper, we propose to resolve such a contradiction from the perspective of anti-overfitting. First, we show that the sharpness at the mini-batch level approximately equals the sum of the gradient norm and the Kullback–Leibler (KL) divergence between the predictive distributions evaluated at the current point and its adversary. Second, we show that the \emph{gradient consistency}, which approximately quantifies the inner product between the mini-batch gradient and the per-example gradient, decreases consistently during training. This result explains why SAM is less likely to suffer from overfitting. At last, building on these insights, we further introduce CASAM, an algorithm that reweights each example according to gradient consistency to enhance the generalization performance. Extensive experiments on popular benchmarks such as ImageNet-1K and Clothing1M corroborate its efficacy.

Existing 3D scene-grounded Large Language Models (3D-LLMs) focus on answering questions grounded in simplified single-room 3D scenes, lacking the ability to reason over real-world household environments containing multiple interconnected rooms and diverse object categories. We introduce CAIRN, a topology-aware 3D-LLM for multi-room 3D scene understanding. CAIRN aligns transformer attention with scene hierarchy, giving the model explicit awareness of object-level relations and room-level connectivity. It enriches object tokens with room-local relational context via a graph neural network, introduces learned room tokens for room-level abstraction, and applies a hierarchical attention mask with geometric bias to route information according to scene topology. CAIRN is developed on CAIRN-MR, a benchmark we introduce on HM3D for multi-room 3D scene understanding, covering grounding, captioning, and four question-answering tasks that progressively evaluate from intra-room perception to cross-room reasoning. Experiments show that CAIRN outperforms prior 3D-LLMs by a large margin across all CAIRN-MR tasks while remaining competitive on five single-room benchmarks.

Most coding-agent benchmarks ask whether generated code behaves correctly. That remains essential, but repository-level engineering is increasingly agent-managed: one agent writes a repository, and later agents inspect, audit, or extend it as working context. In that setting, a generated repository is not only an answer to a task but also a communication artifact for future work. Even when strong agents nearly satisfy the visible behavioral objective, repositories can differ in how clearly they expose the intended behavior and design choices behind that behavior. We introduce Build-and-Find, a protocol for evaluating whether downstream agents can recover those intended choices from generated repositories, and how much inspection that recovery requires. For each task, a builder sees a hidden repository specification and creates a codebase; a finder sees only the codebase and a specification-traced multiple-choice question bank. The protocol separates behavioral correctness from artifact-side recovery and reports recovery accuracy, repeatability, implementation coverage, and inspection effort. Accuracy and stability act as gates: effort is interpreted only when recovery succeeds reliably. Among artifacts from which the same intent can be recovered, lower effort by the same finder suggests that the artifact makes that intent easier to locate. Question-only and spec-only controls quantify generic priors and specification access, while audits separate omitted claims from finder failures and check whether correct answers cite artifact evidence. In the released high-prior task pack, recovery accuracy is near saturation, so inspection effort and finder-specific effects provide the main panel-local comparison. We release the harness, tasks, generated artifacts, records, canonical tables, reports, scripts, metadata, licenses, and evidence audit to support auditable extensions and private task packs.

We study long-context balanced entropic optimal transport (OT) attention on TPU hardware through a stopped-base, fixed-depth tail-refinement surrogate. After a stopped $T$-step Sinkhorn solve, we unroll a short refinement tail and differentiate that surrogate exactly. For the production $R=2$ case, the backward pass contains four staircase plan factors. We prove an exact one-reference-tile schedule: the $R=2$ score cotangent is a single reference plan tile times an explicit modifier field built from vector cotangents and dual differences. This yields block-wise cost $O((T+R)LW)$, $O(Ld)$ input storage, and $O(L)$ additional HBM usage for fixed head dimension $d$ and band width $W$. We also formalize the current \texttt{dustbin\_block} path as the same balanced surrogate on an augmented support, so the schedule lifts to the gap-aware transport path used in our TPU runs. We provide a local surrogate-bias bound, an a posteriori bias certificate, and a projective contraction certificate for strictly positive active blocks. On synthetic masked problems, the optimized kernel matches exact autodiff of the same centered surrogate to within $10^{-5}$--$10^{-10}$. On TPU v6e-8, a four-configuration Pfam screen completes end-to-end, and a promoted balanced $R=2$ run sustains roughly $8.5$ examples per second through a three-hour budget, reaching step $1437$. Held-out Pfam test shards improve reconstruction from $3.17$ to $0.99$ and sparse CE from $5.86$ to $5.69$ relative to step $0$. These results support exact fixed-depth backward theory, a theorem-matching gap-aware bridge, and trainability evidence for the production path.

Vision-language models (VLMs), such as CLIP, have shown remarkable transferability in the biomedical domain, and prompt tuning enables efficient adaptation with limited supervision. In practice, biomedical images exhibit pronounced acquisition-induced style shifts, especially across imaging sites and protocols, while cross-modality transfer further introduces compound shifts from different physical acquisition processes. However, existing prompt-tuning methods ignore acquisition-induced style shifts, making the prompts sensitive to source-specific style and limiting the transferability to unseen medical domains. To address this, we formulate acquisition-induced style shifts as mixture shifts, where each domain is viewed as a mixture of latent style components. Under this formulation, the target risk depends on both mixture weights and latent component risks, motivating prompt adaptation that accounts for style-specific variation rather than relying on a single source-specific prompt. In this paper, we propose SaMoE, a Style-aware Mixture-of-Experts Multimodal Prompt Learning framework for adapting VLMs to biomedical domains under acquisition-induced style shifts. To account for latent component risks in prompt adaptation, SaMoE represents multimodal prompts with latent style-conditioned prompt transformations and composes them through image-conditioned routing, enabling each image to induce style-adapted visual prompts. By extracting style cues from intermediate visual representations, the sparse expert composition adapts style-related prompt weights without domain labels and is injected across multiple VLM layers to align with hierarchical visual-language representations. Extensive experiments on 15 medical datasets across 12 modalities and 10 organs demonstrate significant improvements in both accuracy and generalizability over state-of-the-art methods.

Stereo matchers estimate depth by comparing the left and right images from a stereo camera pair, but they assume that the two cameras are accurately calibrated and rectified. In real deployments, small physical shifts after calibration, caused by camera motion, mechanical drift, vibration, or installation changes, can break this assumption and degrade depth estimation. Online stereo rectification is therefore needed to correct the image pair during operation and recover a stereo geometry where standard stereo matching can work reliably. Recent online rectification methods mainly optimize row alignment, making the left and right views of the same scene point fall on the same image row. We show that this target, while necessary, is insufficient for deployable stereo depth estimation. A rectifier may make a small set of matched points look well row-aligned, but still distort the full images in ways that hurt stereo matching: useful image regions may be cut off, large blank areas may appear, objects may be stretched or squeezed, or image scale may change unevenly across the view. We propose VirtualRect, an online stereo rectification method designed for reliable real-world deployment. Rather than treating rectification as simply making the left and right image rows line up, this method estimates a rectified virtual stereo camera. This makes the rectified images and the camera parameters come from the same geometry, so the images used for matching and the model used for depth computation remain consistent. To improve stability, it further constrains the rectification behavior over the whole image and reduces the influence of geometrically unreliable matches. Experiments on multiple datasets show that, compared with the prior state-of-the-art online rectifier, VirtualRect reduces vertical flow error by 17\% on Carla-Flowguided and 7\% on Semi-Truck Highway.

Model merging has emerged as a promising paradigm for composing large language models directly in weight space, enabling training-free integration of specialized models. However, existing methods rely on parameter-space averaging that systematically induces generation dysregulation---a spectrum of failure modes ranging from structural repetition and termination failure to circular reasoning, and at the extreme, full semantic collapse into incoherent symbols. Even when source models exhibit near-zero dysregulation, existing methods introduce it at $>$97\% (14B/32B scales), collapsing reasoning benchmarks by up to 70 points. We propose Sparse Complementary Fusion with Reverse KL (\textbf{SCF-RKL}), a data-free merging framework that selects complementary parameters via reverse KL divergence on parameter-group proxy distributions. This structure-preserving, sparsity-inducing design maintains proxy-space geometry---theoretically motivated via entropy and subspace bounds---and empirically suppresses generation dysregulation while integrating new capabilities. Extensive experiments on 24 benchmarks across 7B--32B models spanning reasoning, instruction following, safety, and vision demonstrate that SCF-RKL achieves the best overall performance across scales while maintaining near-zero dysregulation ($<$1\%) and strong generalization.

Replay is a standard strategy for mitigating catastrophic forgetting in continual learning, where a bounded memory buffer stores exemplars from previously seen classes. Existing replay methods have largely focused on instance-level decisions, such as which samples to store, replay, or replace, while the class-wise allocation of memory slots is often treated as a uniform heuristic. Here, we study class-wise memory allocation as a complementary design axis and propose VAMA, a value-aware memory allocation framework. VAMA keeps cross-task budgets balanced, estimates task-local class values from Shapley-inspired sample utility scores, and converts them into class-wise quotas through a regularized allocation rule that conservatively deviates from class-balanced replay. This design changes only the class-wise budgeting rule and leaves within-class exemplar selection unchanged, making VAMA easy to integrate into replay pipelines. Experiments on CIFAR-100 and ImageNet-100 show that VAMA consistently improves ER, iCaRL, and FOSTER across different memory budgets and task protocols. Further analyses show that the improvements are not due to non-uniform allocation alone, and that allocation-relevant class-value rankings stabilize early in training, enabling efficient value-aware allocation. These results suggest that class-wise memory allocation is a consequential design choice in replay-based continual learning.

Online map estimation is a crucial component of autonomous driving systems that reduces the reliance on costly high-definition maps. State-of-the-art (SOTA) methods commonly predict map elements as ordered sequences of points that form polylines and polygons. The evaluation of these methods relies predominantly on mean average precision (mAP) based on thresholded Chamfer distance (CD). This framework lacks sensitivity to point ordering and provides limited granularity in assessing geometric quality, making it difficult to distinguish which methods truly excel over others. In this work, we address these limitations on two fronts. For the single-instance similarity measure, we introduce sequence optimal sub-pattern assignment (SOSPA), an order-aware metric that enables fine-grained evaluation of individual geometries while satisfying all metric axioms. For the multi-instance evaluation framework, we propose polyline localisation and detection (PLD), a soft metric that jointly captures detection quality and geometric accuracy, replacing the hard thresholding of mAP with a principled soft assignment. Through evaluations on nuScenes, we demonstrate that PLD effectively ranks SOTA online mapping methods (MapTRv2, StreamMapNet, MapTracker) while providing a decomposed error analysis. This analysis identifies detection capability as the dominant bottleneck in current methods, revealing a performance trend that mAP fails to capture. Code for evaluation using our metrics will be released.

We study fixed-confidence best arm identification in generalized linear bandits under a hybrid feedback model: at each round, the learner may query either (i) absolute reward feedback from a single arm or (ii) relative (dueling) feedback from an arm pair, both governed by generalized linear models. We introduce a likelihood-ratio–based confidence sequence that unifies heterogeneous generalized linear observations and yields an explicit ellipsoidal confidence set under a self-concordance assumption. Building on this confidence set, we propose a hybrid Track-and-Stop algorithm that adaptively allocates queries by tracking a minimax-optimal design over a joint action space of arms and pairs. We establish $\delta$-correctness and provide high-probability upper bounds on the stopping time. We further extend the framework to a cost-aware setting that accounts for heterogeneous acquisition costs across feedback modalities. Empirical experiments demonstrate that the proposed algorithms significantly improve sample efficiency over baseline methods.

Facial behaviour analysis requires understanding valence–arousal (VA), expressions (EXP), action units (AUs), yet existing systems treat these tasks independently or use naïve MTL, ignoring their structured dependencies. This leads to negative transfer, conflicting gradients, poor cross-db generalisation, especially with heterogeneous datasets with only partial labels. We propose Behaviour4All, the first dependency-aware toolkit that unifies VA, EXP, AUs via explicit relational priors and principled optimisation framework. Behaviour4All integrates psychophysically grounded task mappings with coordinated loss families to mitigate gradient conflicts and enforce coherent cross-task semantics. Across eight in-the-wild datasets, Behaviour4All achieves sota performance, improves fairness and generalisation, and outperforms vanilla MTL and student--teacher baselines. We will release the complete toolkit (code, pre-trained models, etc) upon acceptance.

Geometry Problem Solving (GPS) serves as a rigorous touchstone for evaluating the complex reasoning capabilities of Multimodal Large Language Models (MLLMs), particularly when solutions necessitate auxiliary constructions absent in the visual input. However, current approaches face the following challenges: explicit diagramming methods suffer from unstable image generation quality and error cascading, while existing implicit attention mechanisms are restricted to visible elements, failing to perceive the latent auxiliary structures required for deduction. To bridge this gap, we propose AuxCA (Auxiliary-Clues-Aware GRPO), a novel framework that internalizes the capability of auxiliary construction into the model’s reasoning intuition without relying on external tools. Specifically, we introduce Auxiliary Clues Dependency (ACD) to quantify the causal influence of visual cues and integrate it into GRPO via a fine-grained advantage weighting strategy. Complementing this, we incorporate an auxiliary incentive reward to explicitly encourage the model to actively attend to and utilize these geometric elements during reasoning. Extensive experiments demonstrate that our resulting model, AuxCA-8B, significantly outperforms state-of-the-art explicit plotting methods and achieves performance competitive with top-tier proprietary models. Code is available in \url{https://anonymous.4open.science/r/AuxCA-5510}.

Link prediction (LP) is a widely applied task in graph learning. Conventional LP approaches typically follow a dataset-specific paradigm, requiring independent training for each graph and incurring high computational and maintenance costs in large-scale applications. Motivated by these limitations, recent work explores Universal Link Prediction (ULP), aiming to enable training-free inference on arbitrary unseen graphs. However, existing methods primarily rely on subgraph sampling strategies to construct universal representations, which are unable to capture higher-order information due to the exponential growth of higher-order neighborhoods. As a result, they often under-perform on graphs where long-range signals are essential, such as biological networks. Moreover, their prediction models are typically domain-agnostic and lack the ability to adapt to different graph domains, leading to limited generalization. To address these challenges, we propose an Atlas-based Universal Link Prediction framework (AtlasULP), which introduces the relation atlas to capture structural associations between node pairs from both global and local perspectives, enabling a comprehensive characterization of connectivity signals across diverse graph domains. Building upon this representation, we develop a domain-guided prompting ULP model, which generates domain-aware prompt tokens from contextual structures and performs prompt-guided in-context prediction for adaptive link inference. Extensive experiments demonstrate that AtlasULP consistently outperforms state-of-the-art methods across diverse graph domains.