NeurIPS 2026 Papers

Multi-turn conversation is the predominant form of interaction with large language models (LLMs), where user queries continuously evolve. However, existing LLM routing methods are primarily designed for single-turn interactions or fixed queries settings, overlooking the dynamic nature of multi-turn conversations and the challenge of delayed rewards, thereby limiting their ability to optimize cumulative performance. To address this challenge, we move from myopic, single-turn selection to long-horizon routing for multi-turn conversation. Accordingly, we propose ConvoRouter, which first performs MCTS to explore conversation branches induced by different LLM selections and collect trajectories with high cumulative rewards. ConvoRouter then learns a lightweight routing policy from search-derived data, augmented with retrieval-based future state approximation, enabling multi-turn routing without online search. Experiments on both open-domain and domain-specific conversation tasks across diverse candidate sets of both open-source and closed-source LLMs demonstrate that ConvoRouter significantly outperforms single LLMs and existing routing baselines in task success rate, while achieving a superior performance-cost trade-off when combined with a cost-aware reward.

Native multimodal models build rich internal representations across visual, multimodal, and language streams, yet their answers are usually read only from the final layer. We study this mismatch between where task information is available and where the model is allowed to read from. We formalize it with the availability--accessibility gap (AAG), whose primary form compares a small-budget internal readout reference with a learned final-only parity head under the same supervised answer space and loss. Across document, chart, diagram, counting, compositional, and broad multimodal reasoning tasks, we find that task-relevant information often appears at internal sites before it becomes accessible to the final output. We then introduce adaptive internal readout (AIR), a small policy that selects a few internal sites from the full model stack. Under matched budgets, AIR closes much of the gap and outperforms static fusion, dense fusion, visual-only readout, and random site controls. Ablations show that the selected sites are functionally important for the learned readout pathway.

Reinforcement learning and contextual bandit algorithms have become increasingly common in sequential decision-making applications. When these methods are deployed in high-stakes domains, there is growing interest not only in learning effective policies, but also in conducting statistical inference for quantities learned under adaptive data collection. However, classical procedures applied naively in these settings can fail: even when estimators are unbiased, their variance becomes path-dependent and as a result may not be asymptotically normal. A growing literature has emerged to ameliorate this problem, but solutions tend to be problem specific and often rely on correct specification of a working model. In this work, we develop a unified framework for constructing asymptotically valid confidence intervals to cover smooth functionals of nonparametric M-estimands under adaptive sampling. Under Neyman orthogonality, we provide two novel methods for performing inference: (1) a self-normalized statistic based on the realized quadratic variation of the influence function and (2) a statistic using a plug-in estimate of the conditional variance based on reweighted influence function increments. Our results allow for flexible nonparametric estimation of nuisance parameters and remain valid under model misspecification. Our theory is supported by a simulation study for a dynamic pricing application which demonstrates that this method can produce asymptotically valid confidence intervals where standard methods fail.

Many inference-time language-model pipelines combine a cheap reward signal with an expensive verifier, such as exact answer checking in mathematical reasoning or hidden-test execution in code generation. We formalize this setting using a learning-theoretic lens as generative active search: a cost-sensitive first-positive search problem in which a policy adaptively samples candidates from an unknown distribution, observes cheap scores, and pays for verifier labels until it finds a positive example. For a fixed prompt, the generator and reward model induce two unknown objects: a distribution over reward scores and a score-conditioned success function. When these quantities are known, we characterize the distribution-aware optimal policy using a dynamic programming approach. In the realistic and practical setting where both the score distribution and success function are unknown, we propose ADAP, a shellwise adaptive generate-rank-verify algorithm that progressively increases the number of sampled responses and top-ranked verifications. Under the monotonicity assumption that higher reward scores are no less likely to pass verification, we show that ADAP achieves expected cost within a constant factor of the distribution-aware optimum. We complement this result with learning-theoretic lower bounds, based on a centered star number, showing that structural assumptions on the score--label relationship are necessary. Experiments on mathematical reasoning and competitive programming validate the predicted advantage over both fixed non-adaptive policies and difficulty-adaptive baselines.

Node classification in multimodal graphs plays an important role in many real-world applications, where nodes are described by multiple modalities such as text and images, and the graph structure captures relations among entities. However, most existing Graph Neural Networks (GNNs) still focus on pairwise connections, overlooking higher-order relational patterns. Recent studies have explored simplicial complexes to capture higher-order interactions and integrated them into GNN frameworks. Motivated by this line of work, we propose Adaptive Gated Simplicial Propagation (AGSP), an end-to-end framework for node classification in multimodal graphs. Specifically, AGSP first introduces a simplicial propagation layer to capture higher-order relations beyond pairwise connections; then it applies an adaptive gating mechanism to balance multi-order topological relations, enhancing the discriminative ability of node classification. Extensive experiments validate that AGSP generally outperforms the state-of-the-art baselines, highlighting the effectiveness of combining multimodal information with higher-order graph topology through adaptive gated fusion.

Personalizing Large Language Models (LLMs) directly on local edge servers is becoming increasingly important for privacy-preserving and context-aware applications. However, this potential is bottlenecked by hardware resources: limited GPU memory permits only a fixed number of train-ready LoRA adapters, while compute constraints enforce sequential fine-tuning updates. This creates a critical challenge: scheduling scarce update opportunities across a dynamic stream of resident tenants to improve overall model quality and scheduling stability. Crucially, standard Multi-Armed Bandit (MAB) algorithms fail to distinguish between low-potential tenant streams and fully saturated adapters, leading to wasteful exploration on tasks that offer little further gain. To address this, we propose WCA-UCB (Windowed Convergence-Aware UCB), a scheduler designed for this piecewise-converging non-stationary environment. By formally modeling fine-tuning as a slot-constrained bandit problem with piecewise-converging rewards, WCA-UCB detects when an adapter has saturated to pause its training and resets stale statistics after tenant replacement or harmful drift. We prove a dynamic regret bound of $\tilde{O}(\log T)$ and validate the system on an edge-like single-GPU LoRA prototype running Qwen2.5-1.5B. Results demonstrate that WCA-UCB reduces model perplexity by $2.6$% to $8.6$% and uses up to $5.8\times$ fewer training-target switches than strong non-stationary baselines. These results highlight the necessity of convergence-aware scheduling for scalable local LLM personalization.

Attribute-Missing Graph Clustering (AMGC) is a critical yet challenging task in real-world applications, in which only a subset of nodes hold complete attributes information while others are partially missing. Though existing feature propagation-based methods can effectively recover missing node attributes, they commonly assume uniform contributions across all nodes, ignoring unreliable nodes that may degrade imputation quality. Furthermore, adopting low-pass graph filters often suffers from the over-smoothing problem and inevitably sacrifices discriminative information. %Besides, traditional contrastive learning pulls all the node embeddings evenly, which might conflict with the rule that intra-cluster nodes should be closer to each other. To address these limitations, we propose \underline{\textbf{A}}daptive \underline{\textbf{F}}eature \underline{\textbf{P}}ropagation (AFP) for attribute-missing graph clustering. Specifically, we first design a reliability-aware feature propagation mechanism that adaptively weights edges based on node importance. Then, we introduce a multi-scale embedding module to capture both local and global structural information. Finally, we develop a topology-aware contrastive loss to enhance clustering consistency. Extensive experiments on benchmark datasets demonstrate the superiority of our method.

Cyclic block coordinate methods are widely used in practice for their simplicity and strong empirical performance. Yet, their theoretical behavior is challenging to explain, and setting their step sizes$-$beyond classical coordinate descent for minimization$-$requires careful tuning or line-search machinery. In this work, we develop $\texttt{ADUCA}$ (Adaptive Delayed-Update Cyclic Algorithm), a cyclic algorithm addressing a broad class of Minty variational inequalities with monotone Lipschitz operators. $\texttt{ADUCA}$ is parameter-free and locally adaptive: it requires no global or block-wise Lipschitz constants, it uses no per-epoch line search, and it adapts to local problem geometry. A key feature of the algorithm is using operator information delayed by a full cycle, which makes the algorithm compatible with parallel and distributed implementations, and attractive due to weakened synchronization requirements across blocks. We prove that $\texttt{ADUCA}$ attains (near) optimal global oracle complexity as a function of target error $\epsilon >0$, scaling with $1/\epsilon$ for monotone operators, or with $\log^2(1/\epsilon)$ for operators that are strongly monotone.

Out-of-distribution (OOD) detection plays a pivotal role in ensuring the reliability and trustworthiness of AI systems. Although existing approaches achieve strong performance by leveraging features, logits, or both, most of them cannot generalize well across domains due to either overlooking critical information in low-level representations from shallow layers or utilizing sub-optimal layer selection strategies. In this paper, we propose Multi-Layer Adaptive Mahalanobis-Cosine Similarity (ML-AMCoS), a method that adaptively integrates cosine similarity with class-conditional Gaussian distributions. ML-AMCoS is designed to capture subtle nuances in feature covariance when intra-class variance is low, while suppressing covariance noise when the variance is high. To optimize the utilization of each layer, we calculate contribution weights based on performance against pseudo-OOD samples, which are generated by cut-mixing in-distribution (ID) training images with the four corners of images from different classes. Extensive experiments across diverse domains demonstrate that our method significantly improves robustness and detection accuracy, achieving an average AUROC of 81.13\% and FPR@95 of 40.12\% on near-OOD detection, outperforming the state-of-the-art methods by 7.25\% and 7.08\%, respectively.

We study convex optimization problems over a compact convex set where projections are expensive but a linear minimization oracle (LMO) is available. We propose the *adaptive conditional gradient sliding method* (AdCGS), a projection-free and line-search-free method that retains Nesterov's acceleration with adaptive stepsizes based on local Lipschitz estimates. AdCGS combines an accelerated outer scheme with an LMO-based inner routine. It reuses gradients across multiple LMO calls to reduce gradient evaluations, while controlling the subproblem inexactness via a prescribed accuracy level coupled with adaptive stepsizes. We prove accelerated rates for convex objective functions, matching projection-based methods, without relying on a projection oracle. For locally strongly convex objective functions, we further establish linear convergence without additional geometric assumptions on the constraint set, such as polytopes or strongly convex sets. Experiments on constrained $\ell_p$ regression, logistic regression, and least-squares problems demonstrate that AdCGS improves over projection-free baselines and provides competitive performance when projections are inexpensive.

Cooperative multi-agent reinforcement learning (MARL) has long faced scalability challenges due to the exponential growth of the state-action space as the number of agents increases. Existing methods typically enhance scalability by filtering out communications with low-relevance agents. However, such filtering often relies on global state and fixed graph distribution, limiting adaptability in large-scale, communication-constrained environments. In this paper, we propose a scalable MARL method, called Adaptive Communication Range PPO (ACR-PPO), that decomposes the decision-making under communication budget constraints as a sequential process: a communication policy first selects each agent’s communication range within a given budget, followed by a behavior policy that conditions actions on the resulting neighborhood observations. More importantly, we provide a theoretical guarantee of monotonic performance improvement under communication budget constraints. Experiments across diverse scenarios demonstrate that ACR-PPO preserves policy performance while significantly reducing communication costs through adaptive range control.

Making calibrated online predictions is a central challenge in modern AI systems. Much of the existing literature focuses on fully adversarial environments where outcomes may be arbitrary, leading to conservative algorithms that can perform suboptimally in more benign settings, such as when outcomes are nearly stationary. This gap raises a natural question: can we design online prediction algorithms whose calibration error automatically adapts to the degree of non-stationarity in the environment, smoothly interpolating between i.i.d. and adversarial regimes? We answer this question in the affirmative and develop a suite of algorithms that achieve adaptive calibration guarantees under multiple calibration measures. Specifically, with $T$ being the number of rounds and $C\in[0,T]$ being an unknown non-stationary measure defined as the minimal $\ell_1$ deviation of the mean outcomes, our algorithms attain $\widetilde{O}(\sqrt{T}+(TC)^{\frac{1}{3}})$ for $\ell_1$ calibration error and $\widetilde{O}((1+C)^{\frac{1}{3}})$ for both $\ell_2$ and pseudo KL calibration error. These bounds match the optimal rates in the stationary case ($C=0$) and recover known guarantees in the fully adversarial regime ($C=T$). Our approach builds on and extends prior work [Hu et al., 2025, Luo et al., 2025], introducing an epoch-based scheduling together with a novel non-uniform partition of the prediction space that allocates finer resolution near the underlying ground truth.

LM-based agents process external content, exposing them to prompt injection and multi-turn manipulation. Most safety benchmarks evaluate defenders against fixed attack pools collected before evaluation, single-turn or multi-turn. We present a 21-scenario benchmark for adaptive multi-round attacks against memoryless LLM defenders: an autonomous LLM attacker observes prior defender responses and pivots across rounds, while each defender response is evaluated as a fresh interaction. Holding the 21 scenarios, attackers, defenders, and structured-output scoring fixed, restricting scoring to the first attacker turn yields $0$--$1%$ attack success rate (ASR); allowing 15 rounds of adaptive attack yields $8.3$--$19.0%$. Pooling three frontier attacker LLMs uncovers up to $2.4\times$ as many unique successful attacks as the best single attacker, and the generated attacks have low cosine similarity ($0.02$--$0.14$) to attacks in existing benchmarks. Claude Opus 4.6 and GPT-5.4 are aggregate-equivalent ($8.3%$ vs. $11.1%$ ASR), but their weaknesses differ sharply: on one scenario Opus reaches $60%$ ASR ($95%$ CI $36$--$80%$) while GPT-5.4 and Gemini each stay at $7%$ (CI $1$--$30%$). $15$ of $21$ scenarios distinguish at least one defender pair, yet rankings disagree across scenarios (Kendall's $W = 0.21$). We release the benchmark---21 evaluation scenarios, 10 public development scenarios, the orchestrator, baseline harnesses, and a multi-attacker CLI---plus 945 transcripts from the $3\times3$ frontier matrix, an attack-replay dataset, and $18{,}422$ battles from an open competition's final scoring rounds.

Zero-sum games are the textbook model of strategic competition—yet they can fail to capture conflict even in simple settings. *Harmonic games* provide a more robust framework for opposed interests: they are invariant under strategic equivalence, and naturally extend beyond pairwise competition. This generality comes at a structural cost. We show that, in odd-dimensional harmonic games, the set of totally mixed Nash equilibria is a generically non-convex real algebraic variety of dimension at least one, extending to the boundary of the strategy space. As a result, learning becomes delicate: although finely tuned methods are known, whether adaptive, parameter-agnostic learning is possible has remained open. We resolve this question in the affirmative. We introduce a flexible extrapolation-based variant of follow-the-regularized-leader (FTRL+) that converges to Nash equilibrium while achieving order-optimal regret: $\mathcal{O}(1)$ in self-play and $\mathcal{O}(\sqrt{T})$ against arbitrary opponents. The method is fully adaptive—each player updates from locally observable information alone, without access to global problem parameters or shared signals—yielding the first parameter-agnostic learning dynamics provably convergent in this class.

Test-Time Adaptation (TTA) for black-box models accessible only via APIs remains a largely unexplored challenge. Existing approaches such as post-hoc output refinement offer limited adaptive capacity, while Zeroth-Order Optimization (ZOO) enables input-space adaptation but faces high query costs and optimization challenges in the unsupervised TTA setting. We introduce BETA (Black-box Efficient Test-time Adaptation), a framework that addresses these limitations by employing a lightweight, local white-box steering model to create a tractable gradient pathway. Through a prediction harmonization technique combined with consistency regularization and prompt learning-oriented filtering, BETA enables stable adaptation with no additional API calls and negligible latency beyond standard inference. BETA achieves a +7.1% accuracy gain on ViT-B/16 (ImageNet-C) and +3.4% on CLIP (ImageNet-S/R), surpassing strong white-box and gray-box methods including TENT and TPT. On a commercial API, BETA achieves comparable performance to ZOO at 250x lower cost while maintaining real-time inference speed, establishing it as a practical and efficient solution for real-world black-box TTA.

Organoids are an increasingly central scientific platform for drug screening, disease modeling, and regenerative medicine. Because organoid form is closely tied to function, AI-driven image analysis has drawn intense interest, yet every existing organoid imaging study evaluates representations on a single dataset and no work has quantitatively compared representations across the breadth of public organoid datasets. We close this gap with \textbf{OrgBench}, a benchmark of 12 vision foundation models (VFMs) on 13 public organoid-imaging datasets spanning detection, segmentation, classification, and temporal outcome prediction. The benchmark reveals a structured domain gap: task-family winners shift across the benchmark and parameter count is a poor predictor of transfer. Building on this, we introduce \textbf{OrgFM}, a small bottleneck adapter that we continually pre-train on unlabeled organoid images while keeping a general-purpose vision transformer frozen. On the OrgBench task-family aggregates, OrgFM improves over its frozen base backbone. Layer- and feature-level interventions localize the gains to a small set of late-layer morphology-sensitive features: the adapter sparsely supplements, rather than replaces, the base representation. Together, this is the first cross-task, cross-backbone study of organoid imaging at scale; the OrgBench/OrgFM pairing establishes that the parameter-efficient adapter direction is a promising path for organoid representations and clarifies why it works.

Rigorous uncertainty quantification is essential for the safe deployment of autonomous systems in unconstrained environments. Conformal Prediction (CP) provides a distribution-free framework for this task, yet its standard formulations rely on exchangeability assumptions that are violated by the distribution shifts inherent in real-world robotics. Existing online CP methods maintain target coverage by adaptively scaling the conformal threshold, but typically employ a static nonconformity score function. We show that this fixed geometry leads to highly conservative, volume-inefficient prediction regions when environments undergo structural shifts. To address this, we propose \textbf{AdaptNC}, a framework for the joint online adaptation of both the nonconformity score parameters and the conformal threshold. AdaptNC leverages an adaptive reweighting scheme to optimize score functions, and introduces a replay buffer mechanism to mitigate the coverage instability that occurs during score transitions. We evaluate AdaptNC on diverse robotic benchmarks involving multi-agent policy changes, environmental changes and sensor degradation. Our results demonstrate that AdaptNC significantly reduces prediction region volume compared to state-of-the-art threshold-only baselines while maintaining target coverage levels.

Offline goal-conditioned reinforcement learning (GCRL) aims to learn a single policy that adapts its behavior to arbitrary goals from a fixed dataset, making injection of the goal information into the policy a central design question. While recent works increasingly leverage expressive conditional generative modeling architectures, which have demonstrated remarkable success in computer vision, architectural choices for goal conditioning in offline GCRL remain under-explored. We identify that goals in GCRL differ from typical conditioning signals in two ways: they are often noisy or redundant, and they are consumed within a time-sequential episode where the relevant context of a fixed goal shifts as the agent's state evolves. Motivated by these observations, we propose State-Anchored Goal Conditioning (SAGC), a modulation-based goal conditioning scheme that derives its scale and shift parameters by jointly processing state and goal representations in a learned latent space. We further introduce AdaptFlow, an end-to-end trainable offline GCRL architecture that integrates conditional flow matching with SAGC. AdaptFlow outperforms matches all baselines on 23 out of 27 OGBench tasks.

Spherical Transformers have emerged as a promising framework for panoramic semantic segmentation (PASS) by operating directly on spherical geometry and alleviating projection-induced distortions. However, existing architectures rely on assumptions of canonical spherical structure and stable viewpoints, which are frequently violated in real-world $360^\circ$ imagery due to unconstrained camera motion, introducing significant contextual and geometric ambiguity. Consequently, they lack adaptive mechanisms to model such ambiguity, limiting robustness to unseen spherical transformations. In contrast, biological perception is inherently ambiguity-aware: rather than estimating uncertainty probabilistically, it adapts to fluctuations in cue reliability caused by geometric and contextual variations, enabling stable interpretation under complex visual transformations. Motivated by these observations, we first present a systematic analysis of existing PASS architectures under various unseen spherical transformations. Following this, we introduce AdapToPASS, a novel, bio-inspired Spherical Transformer that adaptively models contextual and geometric ambiguities for robust panoramic semantic segmentation. At the core of AdapToPASS are the Adaptive Spherical Attention (AdaSpA) blocks that dynamically modulate attention according to local contextual ambiguity, mimicking the adaptive, context-driven perception of biological vision. To address geometric ambiguities, AdapToPASS employs Bifocal Spherical Representation that reconciles the trade-off between field of view and spatial resolution; along with boundary supervision to emulate the boundary-sensitive nature of biological vision. We evaluate AdapToPASS in both indoor and outdoor semantic segmentation, where it consistently outperforms prior state-of-the-art methods. We further validate AdapToPASS under unseen spherical transformations, where it demonstrates strong robustness and surpasses the next-best method by +13.38% relative improvement in mIoU on Stanford2D3D and +18.77% on WildPASS. Additionally, we introduce a lightweight variant, AdapToPASS-Tiny, with fewer than 2M parameters, which surpasses compact baselines while retaining robustness to spherical transformations.