NeurIPS 2026 Papers

While parameter-efficient fine-tuning methods like low-rank adaptation (LoRA) are standard for large language models, principled estimation of epistemic uncertainty remains challenging. Recent results in the LoRA regime suggest that discrete multi-mode approaches such as deep ensembles offer little benefit over single-mode methods. This contradicts broader observations in deep learning, where ensembling independent optima typically improves generalization, and linking these modes through continuous low-loss valleys further enhances Bayesian model averaging (BMA). Whether such structure exists in the LoRA space and whether it yields functional diversity missed by local or discrete methods has not been studied. We introduce *LoRA-Curve*, a segmented Bézier curve parameterization in the LoRA space, with two variants: a free configuration that jointly optimizes all control points, and an anchored configuration that connects independently fine-tuned LoRA optima. We prove pathwise continuity and Lipschitz regularity of the loss along the curve and empirically show, across reasoning and classification benchmarks with Qwen2.5 7B, that linear interpolation encounters loss barriers, while our anchored multi-segment curves connect independent optima through continuous low-loss valleys. Combined with flat-minima perturbations and a Jensen-Shannon divergence regularizer, *LoRA-Curve* yields measurably higher mutual information of the predictive distribution without sacrificing performance, and links continuous parameter-space traversal to functional diversity.

This work proposes a framework for public decision processes with a monetary component, where voters can pledge donations to support preferred projects and influence the allocation of public funds. It captures a range of real-world scenarios, including participatory budgeting, blockchain-based funding platforms, charitable programs, and crowdfunding campaigns. Through both theoretical and experimental evaluation, our study addresses questions concerning the existence, structure, number, computation, and quality of Nash equilibria.

We introduce a refined differentially private (DP) data structure for kernel density estimation (KDE) with $\ell_1$, $\ell_2$ and $\ell_p^p$ kernels. This new DP data structure offers not only improved privacy-utility tradeoff but also better query efficiency over prior results. Specifically, we study the mathematical problem: given a similarity function $f$ (or DP KDE) and a private dataset $X \subset \mathbb{R}^d$, our goal is to preprocess $X$ so that for any query $y \in \mathbb{R}^d$, we approximate $\sum_{x \in X} f(x, y)$ in a differentially private fashion. The best previous algorithm for $f(x, y) = |x - y|_1$ is the node-contaminated balanced binary tree by [Backurs, Lin, Mahabadi, Silwal, and Tarnawski, ICLR 2024]. Their algorithm requires $O(nd)$ space and time for preprocessing with $n = |X|$. For any query point, the query time is $\alpha^{-1} d \log^2 n$, with an error guarantee of $(1+\alpha)$-approximation and $\varepsilon^{-1} \alpha^{-0.5} d^{1.5} R \log^{1.5} n$. In this paper, we use the same space and pre-processing time, improve the best previous result [Backurs, Lin, Mahabadi, Silwal, and Tarnawski, ICLR 2024] in three aspects - We reduce query time by $\alpha^{-1} \log n$ factor - We improve the approximation ratio from $\alpha$ to $1$ - We reduce the error dependence by a factor of $\alpha^{-0.5}$ From a technical perspective, our method of constructing the search tree differs from previous work [Backurs, Lin, Mahabadi, Silwal, and Tarnawski, ICLR 2024]. In prior work, for each query, the answer is split into $\alpha^{-1} \log n$ numbers, each derived from the summation of $\log n$ values in interval tree countings. In contrast, we construct the tree differently, splitting the answer into $\log n$ numbers, where each is a smart combination of two distance values, two counting values, and $y$ itself. We believe our tree structure may be of independent interest.

Traffic scene understanding requires models to reason beyond object recognition, including lane topology, multi-view geometry, temporal evolution, and signal-phase semantics. However, existing traffic-oriented multimodal benchmarks largely emphasize passive visual recognition or isolated video understanding, offering limited support for evaluating structure-aware traffic reasoning under controlled conditions. We introduce \textbf{OmniTraffic}, a controllable generation pipeline and benchmark for spatio-temporal traffic reasoning. Built around 12 real-world intersections reconstructed into editable 3D traffic environments and complemented by surveillance footage from two countries, OmniTraffic supports both controlled and natural-condition evaluation. It defines a three-level task hierarchy spanning scene perception, multi-view and temporal reasoning, and decision support. Using structured traffic metadata, OmniTraffic generates synchronized multi-view VQA samples covering vehicle states, lane functions, view--BEV correspondence, temporal dynamics, and signal-phase analysis, resulting in 8M VQA samples and a 3K human-verified test set. Evaluation of eleven frontier MLLMs reveals a large human--model gap, with the most pronounced failures in topology-grounded and spatio-temporal reasoning tasks. Fine-tuning a lightweight MLLM on simulated OmniTraffic data further improves performance on real-world traffic scenes, demonstrating the value of simulation-generated supervision for traffic-specific multimodal reasoning. Beyond a fixed dataset, OmniTraffic provides an extensible pipeline with configurable intersections, camera views, traffic demands, signal phases, visual conditions, and rare events.

Offline reinforcement learning (RL) offers a promising route for developing plasma controllers from historical tokamak data, since online trial-and-error on real devices is costly and risky. However, progress in this direction remains difficult to measure due to the lack of a standardized offline RL benchmark for realistic multi-actuator, long-horizon plasma control problems in nuclear fusion. We introduce RL4F, an Offline Reinforcement Learning Benchmark for Plasma Control in Nuclear Fusion, providing closed-loop evaluation environments and baseline comparisons across four full-profile tracking tasks: rotation, density, temperature, and pressure. The dynamics function underlying the evaluation environment is built from historical discharge data from DIII-D, a real-world Tokamak. We evaluate a broad set of imitation learning and offline RL baselines under a unified protocol. We find that offline model-based RL methods obtain the best average performance on most objectives, although no single method dominates all tasks, highlighting the importance of dynamics modeling in complex, long-horizon plasma control tasks. To foster further research, we open-source the codebase, datasets, and evaluation framework, providing a benchmark not only for the fusion community but also for algorithm development in offline RL. Our data and code are available at https://anonymous.4open.science/r/Anonymous-B71F

Offline evaluation of agentic systems often collapses trajectories to terminal success, discarding information about partial progress and inducing widespread ties. We argue that this trajectory collapse creates substantial statistical inefficiency by reducing effective sample size and weakening the ability to distinguish systems. We propose preference-based trajectory evaluation, which compares trajectories directly through temporal preferences over progress and time-to-return profiles. Across diverse agentic and interactive benchmarks, standard success-based metrics produce tied comparisons on roughly 75\% of instances, whereas trajectory-aware preferences reduce ties to roughly 30\%, improving discriminative power, ranking stability, and data efficiency. Our results suggest that benchmark saturation, often portrayed as the result of poor data collection or problem difficulty, may also be explained by the choice of evaluation measure.

The recent trend of integrating autonomous agents into operating systems, supported by the Model Context Protocol (MCP) tools and Deep Searching, has driven research on efficient assistants for daily office tasks like document writing, spreadsheet analysis, and presentation generation. However, training these agents still faces several challenges: the overemphasis on question-answering tasks at the expense of multi-turn generative tasks; the absence of a training paradigm that jointly enhances coding and tool-calling capabilities; the difficulty of evaluating tool efficiency and artifact quality in the presence of drifting user intents. To address these problems, we deploy a unified multi-turn agent framework to collect real-world user interactions, creating \textit{OfficeInstruct}: a comprehensive dataset of 199k high-quality trajectories across Doc, Spreadsheet, and PPT, with an average of 56.2 steps. To ensure data quality, we propose a graph-based evidence evaluation method that models interaction histories as directed graphs, filtering out redundant tool cycles, execution errors, and hallucinations. Experiments demonstrate that our \textit{OfficeAgent} 8B and 14B models surpass open-source baselines on in-domain and out-of-distribution benchmarks, with the 32B variant attaining a 96.19\% success rate on the newly proposed \textit{OfficeAgentBench}, surpassing proprietary models like Claude-4.5-Sonnet and Gemini-3-Pro.

While object detection has advanced through improved architectures and open-vocabulary models, we provide strong evidence that benchmark quality is limited by annotation incompleteness. Across four widely used datasets (COCO, Pascal VOC, Cityscapes, KITTI), re-annotation reveals substantial increases in annotated objects (e.g., up to +60% on KITTI and +40% on COCO), driven primarily by previously unlabeled small, occluded, or densely packed instances. While some differences arise from dataset-specific annotation conventions, we consistently find that missing annotations are the main source of label errors across all datasets. To achieve high data quality, we introduce a scalable annotation pipeline that emphasizes high recall and captures ambiguity through soft labels aggregated from at least 11 annotators per object. The resulting annotations improve coverage and align well with human calibration. We show that benchmark performance is highly sensitive to annotation quality, although model rankings remain largely stable. We introduce two large-scale benchmarks: (i) an uncertainty-aware object detection benchmark, and (ii) a label error detection benchmark grounded in real label errors. We show that current detectors are strongly depended on annotation quality and are misaligned with human perception. Current label error detection methods, which have been shown to perform well on synthetic noise, struggle to achieve high recall and precision on real label errors. Our results highlight the need for future object detection benchmarks to move beyond deterministic annotations toward high-recall, uncertainty-aware evaluation that maximizes valid instances and better reflects real-world ambiguity.

Recent advances in deep learning for shadow removal have significantly enhanced image quality and realism. However, most approaches rely on real-world paired datasets, which are costly to collect and often limited in scene diversity, leading to limited generalization. To address these limitations, we propose One-step Deshadow Diffusion via Reward guidance (ODDR), a new framework that achieves efficient and high-fidelity shadow removal without relying on real-world paired supervision. Our method begins with One-step Deshadow Diffusion (ODD), a baseline model trained on synthetic shadow data for efficient one-step shadow-free reconstruction. We further adapt ODD into ODDR using ShadowReward. In contrast to traditional, annotation-heavy approaches, ShadowReward is the first reward model for shadow removal trained entirely without human annotation. It learns to mimic human perceptual judgments by ranking synthetically generated images with controlled degradations, such as texture distortion and boundary artifacts. This reward-guided fine-tuning enables ODDR to close the synthetic-to-real domain gap. Extensive experiments show that ODD achieves strong performance without relying on real-world paired supervision, and ODDR further improves the results and achieves performance competitive with methods trained on real-world paired data, while maintaining higher computational efficiency as a single step model.

Retrieval benchmarks are increasingly saturating, but we argue that efficient search is far from a solved problem. We identify a class of queries we call *oblique*, which seek documents that *instantiate a latent pattern*, like finding all tweets that express an implicit stance, chat logs that demonstrate a particular failure mode, or transcripts that match an abstract scenario. We study three mechanisms through which obliqueness may arise and introduce OBLIQ-Bench, a suite of five oblique search problems over real long-tail corpora. OBLIQ-Bench exposes an overlooked asymmetry between retrieval and verification, where reasoning LLMs reliably recognize latent relevance whenever relevant documents are surfaced, but even sophisticated retrieval pipelines fail to surface most relevant documents in the first place. We hope that OBLIQ-Bench will drive research into retrieval architectures that efficiently capture latent patterns and implicit signals in large corpora.

Sequential Locate-and-Edit (L\&E) model editing can fail abruptly after many edits. We identify and formalize this failure as a positive _norm-feedback loop_, in which solved value vectors and edited MLP weights progressively amplify each other, degrading edit quality and eventually collapsing model capabilities. Our analysis shows that this feedback can yield approximately exponential norm growth under standard L\&E dynamics, and can remain unresolved by existing increment-level regularizers or update clamps. We propose $\textbf{Norm-Anchor Scaling (NAS)}$, a plug-in stabilizer that breaks this loop by rescaling each solved value vector to an original-model reference norm. Across multiple LLM backbones, datasets, and L\&E editors, NAS extends the usable editing horizon by more than $\textbf{4$\times$}$ and improves long-run editing performance by $\textbf{72.2\%}$ on average, while preserving single-edit efficacy, with only a **one-line modification and negligible computational overhead**. The code is available in the supplementary materials.

We consider a data seller who designs pricing mechanisms over multiple datasets to maximize revenue from budget-constrained buyers. The seller offers multiple datasets and assigns each a pricing function that maps the quantity purchased to a total payment. The goal is to design these pricing functions to maximize revenue, anticipating that buyers—who trade off accuracy gains against cost—choose bundles optimally subject to their budget constraints. Prior work [Chaudhury et al., 2026] studies such optimal pricing under the restriction that each dataset is assigned a linear price, and shows that computing optimal linear prices is computationally intractable. In contrast, we allow each dataset to be priced via a general function and show that this additional flexibility can not only increase the revenue but also restore tractability, yielding a surprising simultaneous improvement in economic performance and computational efficiency. Even when pricing functions are only required to be monotone and lower-continuous, optimal pricing admits a highly structured and simple form: each pricing function is piecewise linear and convex (PLC), and the optimal solution can be computed in polynomial time. Moreover, the total number of kinks across all pricing functions is bounded by the number of buyers. Consequently, when datasets significantly outnumber buyers, most pricing functions are effectively linear. We further empirically study the structure of optimal pricing by analyzing the number of kinks and the revenue gap between optimal nonlinear pricing and optimal linear pricing on simulations generated from a real dataset.

Digital entities such as AI agents and humanoid robots increasingly operate alongside real humans, yet their identity infrastructure is based on credentials rather than embodied biometric identity. We introduce Biometric Identity Provisioning (BIP), a new problem and solution framework that addresses: given an enrollment gallery of real human identities, provision virtual identities that are non-colliding with every enrolled identity, maintain sufficient inter-class separability, and are realizable as high-fidelity face images. The key geometric insight is that real face identities occupy a low-dimensional subspace of the embedding hypersphere, leaving no residual subspace for virtual identities. Hence, virtual identities must instead be allocated as unclaimed gaps within the real face manifold itself. BIP is therefore a constrained packing problem: available gaps vastly exceed any foreseeable enrollment scale, and provisioned identities remain non-colliding even as new real identities are subsequently enrolled. Grounded in this geometry, our repulsion-based allocation is not bounded by any fixed provisioning count; we demonstrate 10M non-colliding virtual identity embeddings against a gallery of 360K real identities. Realizing these embeddings as face images requires a generator that operates outside the training distribution of real face images; we introduce GapGen, a gap-aware generator trained with a curriculum that progressively extends synthesis into non-colliding regions, validated at 1M photorealistic virtual face images. We further construct v-LFW, a virtual counterpart to LFW face dataset, with protocols for virtual face verification, cross-reality matching, real-vs-virtual detection, and unified recognition and detection.

We construct the first coreset of size near-linear in $k$ for fair \kMedian in general metric spaces. Combined with the standard merge-and-reduce framework, our construction also gives the first streaming algorithm for fair \kMedian with space complexity near-linear in $k$. The main technical innovation is a distributional reinterpretation of capacitated clustering that the cost of assigning a dataset to a capacitated center set can be expressed as the earth mover's distance between two discrete distributions. This viewpoint enables us to use tools from optimal transport to analyse uniform sampling. A key ingredient in this analysis is a new family of problem-specific $\eps$-nets for the earth mover's distance.

Functional data such as time series, trajectories, and fields are increasingly central in machine learning and are naturally modeled in infinite-dimensional Hilbert spaces. We study Semi-dual Neural Optimal Transport (SNOT) for learning transport maps between probability measures on separable Hilbert spaces. A critical challenge in this setting is the spurious solution problem: a neural map can globally optimize the SNOT objective while failing to recover an optimal transport map. We show that this failure arises from non-regular source measures, which make the inner minimization in the semi-dual objective non-identifiable. We prove that regular source measures, defined via Gaussian null sets, restore inner-minimizer uniqueness and recover the Monge map. For singular sources, we introduce Gaussian smoothing and establish a necessary and sufficient kernel condition for smoothing to restore regularity. We further prove plan-level consistency: as smoothing vanishes, every Wasserstein accumulation point of the induced optimal plans solves the original Kantorovich problem. Experiments on synthetic singular transports and real-world unpaired time-series imputation benchmarks validate the theory, achieving the best performance on several benchmarks and competitive performance on the remaining ones.

We present Neural Dual Bounds, a calibration-free framework that produces deterministic upper bounds for MAP inference in probabilistic graphical models using a single neural-network forward pass. The framework trains a network to output the edge messages of a mini-bucket style upper bound on a given join-graph decomposition in a self-supervised manner by directly minimizing the a smooth dual objective. The output of this network can be used at test time to warm-start a standard JGLP optimizer. Across diverse benchmarks, this warm start yields dual gaps that are orders of magnitude tighter than those from zero initialization, with the largest gains in the early-iteration regime. We also apply our framework for the constrained MAP (CMPE) task, and we show that neural warm starts with test-time adaptation consistently achieve the tightest dual gaps across all iteration budgets against an adaptive baseline.

Vision-Language-Action (VLA) models promise generalist robotic agents, yet consolidating skills learned across heterogeneous tasks, environments, and embodiments remains difficult: training data is fragmented across platforms, joint multi-task training suffers from interference, and model merging tends to introduce parameter conflicts or require architectural changes. We instead advocate a paradigm in which skills are first learned independently and then consolidated post hoc. To this end, we propose **NestedVLA**, which replaces parameter aggregation with **parameter generation**: a nested hypernetwork, conditioned on the target task, synthesizes task-adaptive parameters that compose with a frozen base VLA. This formulation enables flexible task-conditioned skill reuse while mitigating interference. Across three RL benchmarks (MetaWorld, ManiSkill, CALVIN), NestedVLA outperforms the strongest model-merging baseline by **41 pp on average**, matches per-task fine-tuning **within 1 pp on ManiSkill**, and substantially closes the gap on CALVIN, pointing toward a scalable route to skill consolidation.

We study stochastic linear bandits with delayed feedback under several delay models and establish near-optimal regret guarantees. Our results identify when delayed linear bandits exhibit the same qualitative behavior as multi-armed bandits (MAB), and when the linear structure creates fundamentally new challenges. Specifically, (1) for *loss-independent delays*, where the delay does not depend on the realized loss (but potentially depends on the arm), we show that delays incur only an additive regret penalty. Under stochastic delays, this penalty scales with the expected delay, while under adversarial delays, it scales with the maximum number of outstanding observations. Notably, both delay penalties are dimension-free, improving upon the state-of-the-art results; (2) for *loss-dependent delays*, we show that linear bandits are substantially harder than MAB: unlike in MAB, we prove matching (up to log factors) upper and lower bounds in linear bandits, whose delay penalty depends on the square root of the dimension. (3) for the *delay-as-payoff model*, a special case of loss-dependent delay, we show that the optimal MAB guarantee, which depends only on the delay of the optimal arm, is also unattainable in linear bandits. Together, these results provide a sharp characterization of how delayed feedback interacts with linear generalization.

Robot learning research is fragmented across policy families, benchmark suites, and real robots; Each implementation is entangled with the others in a complex combination matrix, making it an engineering nightmare to port any single element into. General-purpose coding agents may occasionally bridge specific setups, but cannot close this gap at scale because they lack the procedural priors and validation practices that characterize robotics research workflows. We propose NAUTILUS, an open-source harness that turns a single user prompt, for example, ``Evaluate policy $A$ with benchmark $B$''---into ready-to-use reproduction, evaluation, fine-tuning, and deployment workflows. NAUTILUS provides: plug-and-play agent skill sets with distilled priors from robotics research; typed contracts among policies, simulators/benchmarks, and real-world robots; unified interfaces and execution environments; and an agentic coding workflow with explicit, automated validation, and testing at each milestone. NAUTILUS can not only automatically generate the required adapters and containers for existing implementations, but also wrap and onboard new or user-provided policies, simulators/benchmarks, and robots, all connected via a uniform interface. This expands cross-validation coverage without hand-written glue code. Like a nautilus shell that grows by adding chambers, NAUTILUS scales by extending its execution in chambered units, making it a research harness for scalability rather than a hand-curated framework, and aiming to reduce the engineering burden of cross-family reproduction and evaluation in the ever-growing robot learning ecosystem.