# Research Papers — 2026-05 · bundle page 239 of 326

> Full-text markdown of papers #11901–11950 of 16266 published in 2026-05, newest first. AI-readable mirrors of the full 16266-paper 2026-05 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-05/page/239
[← Previous bundle](https://aipod.dev/papers/archive/2026-05/page/238.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-05/page/240.md)

### Test-Time Speculation

Authors: Avinash Kumar, Sujay Sanghavi, Poulami Das
Published: 2026-05-10
Categories: cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2605.09329v2

Speculative decoding accelerates LLM inference by using a fast draft model to generate tokens and a more accurate target model to verify them. Its performance depends on the $\textit{acceptance length}$, or number of draft tokens accepted by the target. Our studies show that the acceptance length of even state-of-the-art speculators, like DFlash, EAGLE-3 and PARD degrade with generation length, reaching values close to 1 (i.e. no speedup) within just a few thousand output tokens, making speculators ineffective for long-response tasks. Acceptance lengths decline because most speculators are trained offline on short sequences, but are forced to match the target model on much longer outputs at inference, well beyond their training distribution. To address this issue, we propose $\textit{Test-Time Speculation (TTS)}$, an online distillation approach that continuously adapts the speculator at test-time. TTS leverages the key insight that the token verification step already invokes the target model for each draft token, providing the training signal needed to adapt the draft at no additional cost. Treating the draft as the student and the target as a teacher, TTS adjusts the draft over several speculation rounds, with each update improving the draft's accuracy as generation proceeds. Our results across multiple models from the Qwen-3, Qwen-3.5, and Llama3.1 families show that TTS improves acceptance lengths over state-of-the-art speculators by up to $72\%$ and $41\%$ on average, with the benefits scaling with increased generation lengths.

---

### Noise-Started One-Step Real-World Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement

Authors: Wei Zhu, Kai Zhang, Yu Zheng, Lei Luo, Yong Guo, Jian Yang
Published: 2026-05-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2605.09328v1

Pre-trained text-to-image (T2I) diffusion models have shown strong potential for real-world image super-resolution (Real-ISR), owing to their noise-started generation process that enables realistic texture synthesis and captures the one-to-many nature of super-resolution. However, diffusion-based Real-ISR methods still face a fundamental efficiency-quality trade-off. Multi-step methods generate high-quality results by iteratively denoising random Gaussian noise under LR conditioning, but suffer from slow sampling. Recent one-step methods greatly improve efficiency, yet they typically replace noise-started generation with direct LR-to-HR restoration, which weakens stochasticity and limits realistic detail synthesis. To address this issue, we propose SMFSR, a noise-started one-step Real-ISR framework via LR-conditioned SplitMeanFlow and GAN refinement. SMFSR preserves the random-noise starting point of diffusion models and learns a direct noise-to-HR mapping conditioned on the LR image. To this end, Interval Splitting Consistency distills the multi-step generative trajectory into a single average-velocity prediction, enabling efficient one-step generation. To compensate for the reduced opportunity for progressive refinement, we further introduce a GAN refinement stage, where a DINOv3-based discriminator enhances realistic texture synthesis and variational score distillation aligns the generated outputs with the natural image distribution under a frozen diffusion teacher. Extensive experiments demonstrate that SMFSR achieves state-of-the-art perceptual quality among one-step diffusion-based Real-ISR methods while retaining fast single-step inference.

---

### PGID: Progressive Guided Inversion and Denoising for Robust Watermark Detection

Authors: Minh Quoc Duong, Chun Tong Lei, Chun Pong Lau
Published: 2026-05-10
Categories: cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2605.09319v1

With the proliferation of AI-generated images, digital watermarking has become an essential safeguard for protecting intellectual property and mitigating malicious exploitation. Recent works on semantic watermarking have enabled efficient copyright protection for diffusion models. However, the dependence of semantic watermarking on diffusion inversion for watermark detection creates a critical vulnerability. Imprint removal and forgery attacks exploit this weakness to produce deceptive results. Our analysis reveals that these attacks succeed by displacing watermarked latents into the unwatermarked region, while guiding unwatermarked latents into the watermarked region. Based on that, we propose Progressive Guided Inversion and Denoising (PGID), the first plug-and-play, training-free noise extraction framework designed to defend against both attack strategies. PGID effectively defends by projecting perturbed latents back to the region where they originally belong. The projection is achieved by eliminating intermediate latent deflections and mitigating adversarial perturbations through progressive inversion-denoising cycles. Comprehensive evaluations across multiple schemes demonstrate that PGID successfully restores detection reliability by recovering removed watermarks and identifying forged instances.

---

### Mem-W: Latent Memory-Native GUI Agents

Authors: Guibin Zhang, Yaohui Ling, Fanci Meng, Kun Wang, Shuicheng Yan
Published: 2026-05-10
Categories: cs.CL, cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2605.09317v1

GUI agents are beginning to operate the web, mobile, and desktop as interactive worlds, where successful control depends on carrying forward visual, procedural, and task-level evidence beyond the fleeting present screen. Yet most agents still treat memory as an external, human-readable artifact: histories are summarized, categorized, retrieved, and reinserted as text or structured records before being encoded again by the policy. This creates a mismatch between the representational form in which experience is stored and the latent embedding sequence over which modern GUI policies actually act. We introduce Mem-W, a series of latent-memory-native GUI agents that treat memory as part of the agent's continuous context rather than as an auxiliary symbolic scaffold. Mem-W weaves both historical trajectories (as experiential memory) and in-session segments (as working memory) into compact memory tokens through a shared trajectory-to-latent compressor. These tokens are woven with the current GUI observation and local context into one continuous embedding sequence, allowing the agent to read successes, failures, and unfinished progress through the same machine-native interface. Mem-W is trained with self-distillation and outcome-aware supervision to preserve decision-relevant state while filtering memory toward evidence that truly supports task success. Across four web and mobile navigation benchmarks, Mem-W consistently improves diverse backbones and memory-enhanced baselines, with gains of up to $+30.0$, suggesting that latent-context-native memory can serve as a scalable foundation for long-horizon GUI agency.

---

### Neural Information Causality

Authors: Jeongho Bang, Marcin Pawłowski
Published: 2026-05-10
Categories: quant-ph, cs.AI
arXiv: https://arxiv.org/abs/2605.09316v1

Query-separated computation forces a representation to play an operational role: data are encoded before a query is known, and a later decoder can answer only through the intermediate interface. In this regime the representation functions as a message rather than merely as a feature map. We formalize this observation by embedding information causality (IC) into representation learning, obtaining a framework called neural information causality (Neural-IC). The revised formulation separates two logically distinct statements. First, every query-separated architecture induces a random-access communication experiment and obeys the embedding inequality $I_{\mathrm{N\text{-}RAC}}\le I(\vec a:H,B)$. Second, any independently certified physical capacity bound on the interface, such as a hard $m$-bit alphabet, a finite-precision register, or a power-constrained noisy channel, implies $I_{\mathrm{N\text{-}RAC}}\le C_H$. This separation avoids treating capacity as a post hoc definition and makes Neural-IC an operational diagnostic for query leakage, precision leakage, and episode-specific memory. We also provide an exact one-bit classical RAC benchmark, showing explicitly that the relevant quantum enhancement is not total information beyond the bottleneck, but fair query-conditioned access. For CHSH-type correlation layers, nested Neural-RAC protocols multiply correlation biases across depth; requiring stability of a one-bit bottleneck for arbitrary depth selects the Tsirelson threshold. We extend the analysis to asymmetric seed biases, to multi-capacity finite-depth phase diagrams, and to correlated data via a conditional information score. Controlled simulations, including straight-through binary bottlenecks and deliberately leaky ablations, verify that apparent violations are accounted for by broken query separation or undercounted capacity.

---

### Do Self-Evolving Agents Forget? Capability Degradation and Preservation in Lifelong LLM Agent Adaptation

Authors: Ye Yu, Xiaopeng Yuan, Haibo Jin, Heming Liu, Yaoning Yu, Haohan Wang
Published: 2026-05-10
Categories: cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2605.09315v1

Recent advances in LLM agents enable systems that autonomously refine workflows, accumulate reusable skills, self-train their underlying models, and maintain persistent memory. However, we show that such self-evolution is often non-monotonic: adapting to new task distributions can progressively degrade previously acquired capabilities across all major evolution channels. We identify this phenomenon as \emph{capability erosion under self-evolution} and show that it consistently emerges across workflow, skill, model, and memory evolution. To mitigate this issue, we propose \emph{Capability-Preserving Evolution} (CPE), a general stabilization principle that constrains destructive capability drift during continual adaptation. Across all four evolution dimensions, CPE consistently improves retained capability stability while preserving adaptation performance. For example, in workflow evolution, CPE improves retained simple-task performance from 41.8\% to 52.8\% under GPT-5.1 optimization while simultaneously achieving stronger complex-task adaptation. Our findings suggest that stable long-horizon self-evolving agents require not only acquiring new capabilities, but also explicitly preserving previously learned ones during continual adaptation.

---

### How LLMs Are Persuaded: A Few Attention Heads, Rerouted

Authors: Xiangkun Sun, Lingkai Kong, Aoqi Zhang, Liang Zeng, Tonghan Wang
Published: 2026-05-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2605.09314v2

Large language models can answer a factual question correctly, yet switch to an incorrect answer when exposed to persuasive context. Why this happens inside the model remains unclear. In this work, we study this question through causal interventions in controlled multiple-choice tasks and identify that persuasion mainly acts through a small set of mid-layer attention heads, which we call decision heads. These heads do not compute the answer themselves. Instead, they copy the option they attend to, so changing their attention can directly change the model's answer. We find that this attention is controlled by a one-dimensional option-routing feature in the option-token representations. Persuasive text changes this feature, causing the decision heads to attend to the persuasion target instead of the correct option. We further trace the routing feature to shallower attention heads that read persuasive keywords from the input. Together, these results reveal a causal pathway: shallow heads read persuasive keywords and write the routing feature, which redirects the decision heads' attention toward the persuasion target; the decision heads then copy the selected option, leading to the wrong answer. We verify each step through intervention and find the same mechanism across four open-source model families and in a source-selection task derived from Generative Engine Optimization (GEO). Finally, based on this mechanism, we introduce a low-rank update to the decision heads' key projections to reduce their sensitivity to persuasion.

---

### How LLMs Are Persuaded: A Few Attention Heads, Rerouted

Authors: Xiangkun Sun, Lingkai Kong, Aoqi Zhang, Liang Zeng, Tonghan Wang
Published: 2026-05-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2605.09314v1

Language models can be persuaded to abandon factual knowledge. This vulnerability is central to AI safety, but its internal mechanism remains poorly understood. We uncover a compact causal mechanism for persuasion-induced factual errors. A small set of mid-layer attention heads almost entirely determines the model's answer. These heads write answer options into a low-dimensional polyhedron, with options occupying distinct vertices. Persuasion does not blur belief or merely reduce confidence; it causes a discrete latent jump from the correct-answer vertex to the persuasion-target vertex. We show that decision heads are not reasoning over evidence. Instead, they copy whichever option token their attention selects. Persuasion works by redirecting attention. We isolate a rank-one evidence-routing feature that controls the route. Directly modifying this feature steers the model's choice, and removing it blocks persuasion. We then trace the feature back to a band of shallower attention heads that build it from persuasive keywords in the input. Every step is validated by intervention. This mechanism appears across open-source LLMs and realistic poisoning scenarios such as Generative Engine Optimization, revealing persuasion as a narrow, monitorable circuit.

---

### Attention Sinks in Diffusion Transformers: A Causal Analysis

Authors: Fangzheng Wu, Brian Summa
Published: 2026-05-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2605.09313v3

Attention sinks -- tokens that receive disproportionate attention mass -- are assumed to be functionally important in autoregressive language models, but their role in diffusion transformers remains unclear. We present a causal analysis in text-to-image diffusion, dynamically identifying dominant attention recipients per timestep and suppressing them via paired, training-free interventions on the score and value paths. Across 553 GenEval prompts on Stable Diffusion~3 (with SDXL corroboration), removing these sinks does not degrade text-image alignment (CLIP-T) or preference proxies (ImageReward, HPS-v2) at $k{=}1$; only under stronger interventions ($k\!\geq\!10$) does HPS-v2 exhibit a metric-dependent boundary, while CLIP-T remains robust throughout. The perceptual shifts induced by suppression are nonetheless \emph{sink-specific} -- $\sim\!6\times$ larger than equal-budget random masking -- revealing an empirical dissociation between trajectory-level perturbation and \emph{semantic alignment} in diffusion transformers. \footnote{Code available at https://github.com/wfz666/ICML26-attention-sink.}

---

### Low-Cost Neural Radiance Fields

Authors: Alice Huang, Prathamesh Sonawane, Yashdeep Thorat, Yug Rao
Published: 2026-05-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2605.09312v1

Neural Radiance Fields (NeRF) achieve high-quality novel-view synthesis, but their long training times and reliance on dense input views limit accessibility. We present a comparative study of three accelerated NeRF variants - DS-NeRF, TensoRF, and HashNeRF and explore extensions targeted at the low-compute, low-data regime. First, we add a depth-supervision loss derived from COLMAP keypoints to TensoRF (TensoRF-DS) and evaluate it on the LLFF dataset under reduced view counts. Second, we ablate the feature-decoding MLP of TensoRF and study the effect of input downsampling on PSNR and runtime on the synthetic Lego scene. Third, we propose four architectural variants of the HashNeRF color and density networks, including residual and convolutional designs, and report PSNR/training-time tradeoffs under matched iteration budgets. Under iso-time evaluation, none of our extensions conclusively outperform the published baselines, but the experiments characterize which extensions transfer to constrained settings and surface design questions for future work.

---

### Teaching Molecular Dynamics to a Non-Autoregressive Ionic Transport Predictor

Authors: Jiyeon Kim, Byungju Lee, Won-Yong Shin
Published: 2026-05-10
Categories: cs.LG, cs.AI, physics.atom-ph, physics.chem-ph, physics.comp-ph
arXiv: https://arxiv.org/abs/2605.09311v1

Unlike most static material properties widely studied in the machine learning literature, ionic transport properties are inherently dynamic, making their fast and accurate prediction from static atomic structures challenging. The current standard approach, molecular dynamics (MD) simulations, suffers from prohibitively high computational cost. Recent autoregressive learning-based MD acceleration methods requiring sequential inference remain slow and prone to error accumulation; in contrast, existing non-autoregressive material property prediction models are less accurate because they fail to exploit dynamics. Moreover, existing methods typically benefit from datasets either with or without atomic trajectories, but not both. To overcome these limitations, we propose a non-autoregressive learning framework based on auxiliary modality learning, which treats atomic trajectories as an auxiliary modality during training but does not require them at inference. This enables the predictor to learn dynamics without sequential inference while benefiting from both types of datasets. As a result, our framework achieves over 200 times speedup compared to autoregressive models on the dataset with atomic trajectories while substantially reducing prediction error relative to non-autoregressive benchmarks across both types of datasets. Our code is available at https://github.com/jykim-git/MD.

---

### Beyond ESG Scores: Learning Dynamic Constraints for Sequential Portfolio Optimization

Authors: Xin Li, Yan Ke, Longbing Cao
Published: 2026-05-10
Categories: cs.AI, q-fin.PM
arXiv: https://arxiv.org/abs/2605.09310v1

ESG-aware portfolio optimization is increasingly important for sustainable capital allocation, yet most learning-based methods still operationalize ESG by appending static scores to the policy observation or reward. This creates a mismatch for sequential control: ESG scores are noisy, provider-dependent, low-frequency, and temporally misaligned with sequential portfolio decisions, while financial evidence suggests that ESG is better treated as a portfolio preference, risk-exposure, or hedge dimension than as a robust alpha factor. We propose to impose ESG constraints without modifying the financial policy's observation or reward, using a Multimodal Action-Conditioned Constraint Field (MACF) that learns mechanism-specific ESG costs from point-in-time multimodal evidence and contemplated portfolio transitions. We then introduce MACF-X, a family of optimizer-specific adapters that converts MACF costs and uncertainties into native constrained-optimization interfaces through a shared slack- and uncertainty-aware pressure layer. Across multiple constraint-integration interfaces, MACF-X reduces tail ESG budget pressure while maintaining competitive financial performance. Ablations show that this improvement depends on dynamic evidence inputs and three-head decomposition, while static ESG-score proxies are nearly indistinguishable from score-shuffled noise baselines.

---

### Hierarchical Attention-based Graph Neural Network with Relevance-driven Pruning

Authors: Seungwoo Kum
Published: 2026-05-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2605.09308v1

Graph Neural Networks (GNNs) excel at relational reasoning but face two persistent challenges: the lack of interpretable attribution for heterogeneous node types, and the computational overhead of message passing over large, noisy graphs. We propose the Hierarchical Attention-based Heterogeneous GNN (HA-HeteroGNN), a framework that addresses both issues through a unied explainability-to-pruning pipeline. A two-tier attention mechanism separates sensor-level and context-level computation across 16 node types and 18 edge types, producing per-node relevance scores via an attention-based GNN Explainer without requiring gradient backpropagation. These relevance scores then serve as a principled pruning criterion: removing nodes identied as consistently uninformative yields a 27% reduction in graph edges while simultaneously improving classication accuracy by 2.46.1% across all model variants, challenging the conventional assumption that pruning necessarily trades accuracy for eciency. Experiments on a 50,000-record synthetic dataset spanning 11 report categories demonstrate 97.5% cross-strategy explanation stability and domain consistent sensor attribution, with training-time reductions of up to 43.9% and real-time inference latency of approximately 5860 ms per sample.

---

### Reinforcement Learning Measurement Model

Authors: Wenqian Xu, Feng Ji
Published: 2026-05-10
Categories: stat.ME, cs.HC, cs.LG, stat.CO, stat.ML
arXiv: https://arxiv.org/abs/2605.09305v1

Interactive assessments generate sequential process data that are not well handled by conventional item response models. Existing MDP-based measurement approaches, such as the Markov decision process measurement model (MDP-MM, LaMar, 2018), link action choices to state-action values, but their reliance on person-specific tabular value functions makes them difficult to scale beyond small, fully enumerated tasks. We propose the Reinforcement Learning Measurement Model (RLMM), a measurement framework that decouples person-level choice sensitivity from task-level value representation through a shared parametric action-value function, making estimation more computationally efficient for larger process-data settings. The model combines a Boltzmann choice rule with normalized advantages, a soft Bellman consistency penalty, and a block-coordinate MAP procedure for joint estimation, while also yielding step-level influence diagnostics for identifying behaviorally critical decisions. In peg-solitaire simulations, the RLMM achieved higher estimation accuracy and substantially lower runtime than the original MDP-MM, with advantages increasing as task complexity grew. In AQUALAB gameplay logs, the estimated person parameter was positively associated with cumulative reward, task completion, and behavioral efficiency. These results show that the RLMM extends decision-process-based psychometric models to larger and more behaviorally realistic environments while preserving an interpretable latent trait tied to decision making steps.

---

### Path-Dependent Denoising: A Non-Conservative Field Perspective on Order Collapse in Diffusion Language Models

Authors: Jeonseong Kim
Published: 2026-05-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.09303v1

Diffusion language models (DLMs) offer a structural alternative to autoregressive generation: denoising can update tokens in arbitrary orders or in parallel rather than along a fixed left-to-right chain. In practice, fast DLM decoding remains strongly order-sensitive and often drifts toward autoregressive-like trajectories. We trace this tension to compatibility. At each reverse-time step, a DLM provides local denoising conditionals over the unresolved tokens. Arbitrary-order denoising becomes well defined when these local conditionals compose into order-invariant pseudo-joints. We formalize this view by defining order-induced pseudo-joints and a local denoising circulation: the log-ratio between the two pseudo-joints obtained by swapping a pair of unresolved positions. This circulation is zero under compatible conditionals, and global order gaps decompose into sums of local circulations along adjacent swaps. We further separate incompatibility-driven path dependence from conditional-dependence error in parallel updates and from order-specific estimation error. The resulting framework provides inference-only diagnostics for testing when DLM decoding is genuinely order-free.

---

### Discrete Langevin-Inspired Posterior Sampling

Authors: Chaitanya Amballa, Sattwik Basu, Jorge Vančo Sampedro, Romit Roy Choudhury
Published: 2026-05-10
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2605.09302v1

We study posterior sampling for inverse problems in discrete state spaces using discrete diffusion models as generative priors. While continuous diffusion models have become widely used for inverse problems, their discrete counterparts remain comparatively underexplored. Existing discrete posterior samplers often rely on continuous relaxations of discrete variables, Gibbs-style updates, or mechanisms specialized to particular corruption processes, which can limit scalability or generality. We propose $Δ$LPS, a Discrete Langevin-Inspired Posterior Sampler that uses gradient information to identify promising discrete moves without leaving the discrete state space. The resulting approach enables efficient parallel updates across all token dimensions and is agnostic to the training paradigm of the discrete diffusion prior, including masked and uniform-state diffusion. We evaluate our method on image restoration tasks across MNIST, CIFAR, and FFHQ, as well as spatial mapping, covering linear, nonlinear, and blind inverse problems. Across these settings, we improve over recent discrete diffusion posterior samplers and are competitive with strong continuous diffusion-based inverse solvers. Our results suggest that fully discrete, gradient-informed posterior samplers offer a scalable and general path toward solving inverse problems over discrete representations.

---

### Neural Cluster First, Route Second: Capacitated Vehicle Routing via Differentiable Optimal Transport

Authors: Samuel J. K. Chin, Maximilian Schiffer
Published: 2026-05-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2605.09301v2

The Capacitated Vehicle Routing Problem (CVRP) underpins modern last-mile logistics, where routing decisions recur over the same fixed service area, like a city. In this setting, routing problems share a fixed set of potential customer locations, while active customers and demands vary between instances. We study how this spatial support can be exploited through reusable learned representations and design our method around three symmetries of the symmetric Euclidean CVRP: $E(2)$ transformations, vehicle-route permutations, and tour reversal. We introduce Neural Cluster-First--Route-Second (CFRS), a neural extension of the Fisher--Jaikumar framework that predicts seed-selection scores and customer-to-cluster assignment costs non-autoregressively and respects the three symmetries. A differentiable entropic optimal transport layer provides capacity-aware supervision and guides discrete capacitated assignment, followed by independent traveling salesman subproblems for route recovery. Component ablations show consistent benefits from learned seed selection, while learned assignment costs perform best near the training size and classical FJ costs perform better at larger sizes under exact decoding. On the fixed-support distribution with constant capacity, a model trained on $N=100$ achieves a $3.77\%$ routing gap relative to HGS at $N=1000$ without retraining. A shallow variant with one attention layer in each transformer achieves a $5.08\%$ gap at this scale, with spatial embeddings consistently improving routing quality over raw coordinates. Embedding interpolation further accommodates entirely unseen customer locations without retraining. On standard CVRP benchmarks, a separately trained model achieves a $2.73\%$ routing gap relative to LKH-3 at $N=100$.

---

### Neural Cluster First, Route Second: One-Shot Capacitated Vehicle Routing via Differentiable Optimal Transport

Authors: Samuel J. K. Chin, Maximilian Schiffer
Published: 2026-05-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2605.09301v1

The Capacitated Vehicle Routing Problem (CVRP) underpins modern last-mile logistics. Current Neural Combinatorial Optimization (NCO) methods construct CVRP solutions autoregressively, inheriting sequential decoding bottlenecks, sensitivity to spatial symmetries, and brittle out-of-distribution behavior. We revisit the classical Cluster-First-Route-Second (CFRS) paradigm -- long known to be asymptotically optimal but largely overlooked by NCO -- and argue that it is structurally aligned with the core strengths of deep learning: similarity and assignment over global context, rather than the construction of long sequential tours. We introduce Neural CFRS, the first purely non-autoregressive one-shot neural CFRS framework for the CVRP. It enforces global fleet-capacity constraints end-to-end via a differentiable entropic Optimal Transport layer, producing a continuous transport plan to sparsify an exact capacitated assignment solver. We provide formal theoretical guarantees that our architecture intrinsically abstracts away $E(2)$ spatial, inter-route permutation, and intra-route traversal symmetries. By equipping the framework with a pre-trained spatial vocabulary, we unlock extreme parameter efficiency and zero-shot scaling. Designed primarily for real-world spatial distributions under a constant capacity setting, Neural CFRS scales robustly to out-of-distribution $N=1000$ instances with a < 4% gap -- retaining an approximate 5% gap at this scale even as an ultra-lightweight, single-layer architecture. Furthermore, when deployed out-of-the-box on standard benchmarks, we achieve a highly competitive 2.73% optimality gap on size-100 problems.

---

### LagrangianSplats: Divergence-Free Transport of Gaussian Primitives for Fluid Reconstruction

Authors: Ningxiao Tao, Baoquan Chen, Mengyu Chu
Published: 2026-05-10
Categories: cs.GR, cs.LG
arXiv: https://arxiv.org/abs/2605.09299v1

Reconstructing 3D fluid velocity fields from sparse 2D video observations is a highly ill-posed inverse problem, demanding both transport consistency with observed motion and physical validity under fluid laws. Existing methods typically impose these constraints through soft penalties, often leading to compromised accuracy and convergence issues. We introduce a reconstruction framework that structurally enforces both constraints. Specifically, we parameterize the reconstructed velocity using a continuous Divergence-Free Kernel representation, driving the advection of a Lagrangian 3D Gaussian Splatting representation. This formulation intrinsically guarantees both flow incompressibility and long-range transport coherence by construction. To enable the efficient optimization of such a constrained system, we introduce a novel Sliding Window scheme that propagates gradients over meaningful temporal horizons while maintaining tractable training costs. Experiments on synthetic and real-world datasets demonstrate that our method outperforms state-of-the-art baselines in both transport consistency and physical accuracy, enabling applications such as high-quality re-simulation and flow analysis.

---

### Micro-Defects Expose Macro-Fakes: Detecting AI-Generated Images via Local Distributional Shifts

Authors: Boxuan Zhang, Jianing Zhu, Qifan Wang, Jiang Liu, Ruixiang Tang
Published: 2026-05-10
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2605.09296v1

Recent generative models can produce images that appear highly realistic, raising challenges in distinguishing real and AI-generated images. Yet existing detectors based on pre-trained feature extractors tend to over-rely on global semantics, limiting sensitivity to the critical micro-defects. In this work, we propose Micro-Defects expose Macro-Fakes (MDMF), a local distribution-aware detection framework that amplifies micro-scale statistical irregularities into macro-level distributional discrepancies. To avoid localized forensic cues being diluted by plain aggregation, we introduce a learnable Patch Forensic Signature that projects semantic patch embeddings into a compact forensic latent space. We then use Maximum Mean Discrepancy (MMD) to quantify distributional discrepancies between generated and real images. Our theory-grounded analysis shows that patch-wise modeling yields provably larger discrepancies when localized forensic signals are present in generated images, enabling more reliable separation from real images. Extensive experiments demonstrate that MDMF consistently outperforms baseline detectors across multiple benchmarks, validating its general effectiveness. Project page: https://zbox1005.github.io/MDMF-project/

---

### LEAF-SQL: Level-wise Exploration with Adaptive Fine-graining for Text-to-SQL Skeleton Prediction

Authors: Zhao Tan, Xiping Liu, Qing Shu, Qizhi Wan, Dexi Liu, Changxuan Wan
Published: 2026-05-10
Categories: cs.CL
arXiv: https://arxiv.org/abs/2605.09295v1

Text-to-SQL translates natural language questions into executable SQL queries, enabling intuitive database access for non-experts. While large language models achieve strong performance on Text-to-SQL with prompting, they still struggle with complex queries that involve deeply nested logic or multiple clauses. A widely used approach employs SQL skeletons--intermediate representations of query logic--to streamline generation, but existing methods are limited by their reliance on a single structural hypothesis and lack of progressive reasoning. To overcome these limitations, we propose LEAF-SQL, a novel framework that reframes skeleton prediction as a coarse-to-fine tree search process. LEAF-SQL enables systematic exploration of diverse structural hypotheses with adaptive refinement. Several key techniques are employed in LEAF-SQL: (1) a three-level skeleton hierarchy to guide the search, (2) a Skeleton Formulation Agent to generate diverse candidates, and (3) a Skeleton Evaluation Agent to efficiently prune the search space. This integrated design yields skeleton candidates that are both structurally diverse and granularity-adaptive, providing a stronger foundation for the SQL generation. Extensive experiments show that LEAF-SQL consistently improves the performance of various LLM backbones. On the official hidden test set of the challenging BIRD benchmark, our method achieves 71.6 execution accuracy, which outperforms leading search-based and skeleton-based methods, affirming its effectiveness for complex queries.

---

### Towards Effective Theory of LLMs: A Representation Learning Approach

Authors: Muhammed Ustaomeroglu, Guannan Qu
Published: 2026-05-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2605.09294v2

We propose Representational Effective Theory (RET), a framework for describing large language model computation in terms of learned macrostates rather than microscopic details. RET learns these macrostates from hidden-state trajectories using a BYOL/JEPA-style self-supervised objective, coarse-graining activations into macrovariables that preserve higher-level structure relevant for prediction and interpretation. We evaluate whether these macrovariables are practically relevant for interpretability: RET yields temporally consistent states that reveal ``mental-state'' trajectories of reasoning, capture high-level semantic structure, support early prediction of behavioral outcomes such as sycophancy and alignment faking, while providing causal handles for steering generations toward interpretable computational phases. Together, these results suggest that LLM computation admits useful effective descriptions via RET: high-level, dynamically meaningful variables for interpretation, prediction, and control.

---

### Towards Effective Theory of LLMs: A Representation Learning Approach

Authors: Muhammed Ustaomeroglu, Guannan Qu
Published: 2026-05-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2605.09294v1

We propose Representational Effective Theory (RET), a framework for describing large language model computation in terms of learned macrostates rather than microscopic details. RET learns these macrostates from hidden-state trajectories using a BYOL/JEPA-style self-supervised objective, coarse-graining activations into macrovariables that preserve higher-level structure relevant for prediction and interpretation. We evaluate whether these macrovariables are practically relevant for interpretability: RET yields temporally consistent states that reveal "mental-state" trajectories of reasoning, capture high-level semantic structure, support early prediction of behavioral outcomes such as sycophancy, and provide causal handles for steering generations toward interpretable computational phases. Together, these results suggest that LLM computation admits useful effective descriptions via RET: high-level, dynamically meaningful variables that support interpretation, prediction, and intervention.

---

### Beyond Accuracy: Evaluating Strategy Diversity in LLM Mathematical Reasoning

Authors: Xia Yang, Xuanyi Zhang, Hao Hu, Feng Ji
Published: 2026-05-10
Categories: cs.AI, cs.CY
arXiv: https://arxiv.org/abs/2605.09292v1

Large language models now achieve high final-answer accuracy on mathematical reasoning benchmarks, but accuracy alone does not capture reasoning flexibility. We introduce a strategy-level evaluation framework instantiated on 80 AMC 10/12 and AIME problems with 217 AoPS-derived reference strategy families. Model outputs are annotated for strategy identity, validity, and correctness using dual-AI coding with human adjudication. Across four frontier models, we find a pronounced decoupling between answer accuracy and strategy diversity. Under a single-solution prompt, all models achieve high accuracy (95%-100%), but under a multiple-strategy prompt they recover substantially fewer strategies than the human reference set. Gemini, DeepSeek, GPT, and Claude generate 184, 152, 151, and 110 distinct valid strategies, respectively, with the largest gaps in Geometry and Number Theory. The models collectively produce 50 benchmark-novel valid strategies, indicating both incomplete coverage of human strategies and some capacity for alternative reasoning. A repeated-run robustness check on 20 problems shows diminishing gains in discovered strategies, with the strongest model recovering only 39 of 55 AoPS-reference strategies (71%) after three runs. These findings position strategy diversity as a complementary dimension for evaluating mathematical reasoning beyond answer correctness.

---

### dFlowGRPO: Rate-Aware Policy Optimization for Discrete Flow Models

Authors: Zhengyan Wan, Yidong Ouyang, Panwen Hu, Qiang Sun
Published: 2026-05-10
Categories: cs.LG, stat.AP
arXiv: https://arxiv.org/abs/2605.09291v1

Discrete flow models (DFMs) are a class of flexible generative models for generating discrete data, and diffusion large language models (dLLMs) can be viewed as a special case with a specific choice of mixture path and a masked source distribution. While several recent works have explored reinforcement learning into dLLMs, its application to more general discrete flow models remains underexplored. In this work, we present discrete Flow-GRPO (dFlowGRPO), a unified reinforcement learning framework for discrete flow models that supports a broad family of probability paths and non-masked source distributions. We derive the full trajectory probability for DFMs and formulate denoising as a Markov decision process, enabling dFlowGRPO to incorporate information from both the associated conditional transition rates and the posterior model during reinforcement learning. We apply dFlowGRPO to FUDOKI, a recent multimodal discrete flow model, and evaluate it on both image generation and multimodal understanding tasks. Empirical results show that dFlowGRPO outperforms existing GRPO-type methods for dLLMs on text-to-image generation tasks and achieves performance competitive with continuous flow-based models trained using FlowGRPO, while also demonstrating strong capabilities on understanding tasks.

---

### From Regression to Inference: Meta-Learning Predictors for Neural Architecture Search

Authors: Liping Deng, MingQing Xiao
Published: 2026-05-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.09290v1

Prediction-based approaches are widely used in neural architecture search (NAS), where a predictor estimates the performance of candidate architectures to guide selection. However, existing predictors are typically trained via supervised regression on limited samples, leading to overfitting and poor generalization to unseen architectures. In this work, we propose a fundamentally different formulation that models performance prediction as a conditional function inference problem using a Convolutional Neural Process (ConvNP) with meta-learning capabilities. Instead of fitting a fixed mapping to limited samples, our approach meta-learns to infer performance from partial observations by training with context-target splits across a group of synthesized tasks, explicitly optimizing for generalization under data scarcity and aligning the training procedure with the deployment setting in NAS. We further design simple yet effective meta-features for cell-based architectures and evaluate our method on NAS-Bench-101 and NAS-Bench-201. Extensive experiments show that our approach consistently improves top-K ranking quality and achieves the state-of-the-art architecture selection using limited samples.

---

### MC$^2$: Monte Carlo Correction for Fast Elliptic PDE Solving

Authors: Ethan Hsu, Hong Meng Yam, Ivan Ge
Published: 2026-05-10
Categories: cs.LG, cs.AI, cs.CE, cs.CV, math.NA
arXiv: https://arxiv.org/abs/2605.09288v1

Partial differential equation (PDE) solvers underpin scientific computing, but real-world deployment is bounded by compute. Classical Monte Carlo solvers such as Walk-on-Spheres (WoS) are unbiased and geometry-agnostic but are slow. Learned solvers are fast but biased and brittle under distribution shift. We present \textbf{MC$^2$}, a hybrid WoS-Neural Network (WoS-NN) PDE solver that treats a low-budget Monte Carlo solution as a structured estimator of the true field and learns a single-pass neural correction to recover a high-fidelity solution. MC$^2$ matches the accuracy of solutions using over $1000\times$ more Monte Carlo compute, outperforming all evaluated classical, denoising, and neural-operator baselines. To enable reproducible study of finite-compute PDE solving, we additionally release \textbf{PDEZoo}, the largest standardized elliptic PDE benchmark to date: 2M PDEs spanning five elliptic families and unlimited geometric compositions, with analytic ground truth and multi-budget Monte Carlo trajectories. Together \textbf{MC$^2$} and \textbf{PDEZoo} (1) empirically establish that finite-sample Monte Carlo error is structured, learnable, and correctable in a single forward pass, (2) show that we can solve PDEs $\sim$\textbf{1000x} faster than with just WoS, and (3) provide the evaluation infrastructure the field has so far lacked.

---

### PiCA: Pivot-Based Credit Assignment for Search Agentic Reinforcement Learning

Authors: Dongyi Liu, Yifan Niu, Qinwen Wang, Han Xiao, Jia Li
Published: 2026-05-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2605.09287v2

Large Language Model (LLM)-based search agents trained with reinforcement learning (RL) have significantly improved the performance of knowledge-intensive tasks. However, existing methods encounter critical challenges in long-horizon credit assignment: (i) Reward Sparsity, where models receive only outcome feedback without step-level guidance to differentiate action quality; (ii) Isolated Credit, where credit is assigned to steps independently, failing to capture sequential dependencies; and (iii) Distributional Shift, where rewards are estimated on templates that deviate from the model's natural generative distribution. To address these issues, we propose Pivot-Based Credit Assignment (PiCA), a novel step reward mechanism that reformulates the search trajectory as a sequential process of cumulative search progress. Unlike prior isolated step rewards, PiCA defines process rewards as success probabilities dependent on the historical context based on Potential-Based Reward Shaping (PBRS). This approach identifies pivot steps, which comprise target golden sub-queries and sub-answers derived from historical trajectories, as information peaks that significantly boost the likelihood of a correct final answer. By anchoring these step rewards to the final task objective, PiCA provides dense, pivot-aware and trajectory-dependent guidance while maintaining distributional consistency. Extensive experiments show that PiCA outperforms existing strong baselines across seven knowledge-intensive QA benchmarks, achieving 15.2% and 2.2% improvements for 3B and 7B models. The consistent performance gains across various models show PiCA's robust generalization. The code is available at https://github.com/novdream/PiCA.

---

### BetaEdit: Null-Space Constrained Sequential Model Editing

Authors: Bingqing Liu, Wei Liu, Yuhua Li
Published: 2026-05-10
Categories: cs.CL
arXiv: https://arxiv.org/abs/2605.09285v1

Null-space-based methods have garnered considerable attention in model editing by constraining updates to the null space of the pre-existing knowledge representation, thereby preserving the model's original behavior. However, in practice these methods rely on an approximate null space--leading to knowledge leakage--and further suffer from severe performance degradation during sequential editing. Recent work shows that history-aware editing strategies can empirically mitigate this decline, yet the underlying reason remains unclear. In this paper, we first expose the knowledge leakage inherent in existing null-space approaches and then analyze why history-aware updates effectively preserve both editing performance and general capabilities during long-horizon editing. Building on these insights, we propose BetaEdit, a refined framework that effectively controls the knowledge leakage and integrates history-aware updates into the null-space paradigm. Extensive experiments on three large language models across two standard benchmarks show that BetaEdit consistently outperforms prior methods in the challenging regime of massive-scale sequential editing. Code is available at: https://github.com/lbq8942/BetaEdit.

---

### Semi-Supervised Neural Super-Resolution for Mesh-Based Simulations

Authors: Jiyeon Kim, Youngjoon Hong, Won-Yong Shin
Published: 2026-05-10
Categories: cs.LG, cs.AI, cs.CE, physics.app-ph, physics.comp-ph
arXiv: https://arxiv.org/abs/2605.09284v1

Mesh-based simulations provide high-fidelity solutions to partial differential equations (PDEs), but achieving such accuracy typically requires fine meshes, leading to substantial computational overhead. Super-resolution techniques aim to mitigate this cost by reconstructing high-resolution (HR), high-fidelity solutions from low-cost, low-resolution (LR) counterparts. However, training neural networks for super-resolution often demands large amounts of expensive HR supervision data. To address this challenge, we propose SuperMeshNet, an HR data-efficient super-resolution framework for mesh-based simulations aided by message passing neural networks (MPNNs). At its core, SuperMeshNet introduces complementary learning, a semi-supervised approach that effectively leverages both 1) a small amount of paired LR-HR data and 2) abundant unpaired LR data via two jointly trained, complementary MPNN-based models. Additionally, our model is enriched by inductive biases, which are empirically shown to further improve super-resolution performance. Extensive experiments demonstrate that SuperMeshNet requires 90% less HR data to achieve even lower root mean square error (RMSE) than that of the fully supervised benchmark without the inductive biases. The source code and datasets are available at https://github.com/jykim-git/SuperMeshNet.git.

---

### A Prompt-Aware Structuring Framework for Reliable Reuse of AI-Generated Content in the Agentic Web

Authors: Shusaku Egami, Masahiro Hamasaki
Published: 2026-05-10
Categories: cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2605.09283v1

The evolution of Large Language Models (LLMs) and the software agents built on them (AI agents) marks a turning point in the transition from a human-centric Web to an ``Agentic Web'' driven by AI agents. However, for AI-Generated Content (AIGC), which is expected to dominate the Web, there is currently no mechanism for agents to verify its reliability, reproducibility, or license compliance during generation. This lack of transparency risks causing chained hallucinations and compliance violations through the reuse of AIGC. Consequently, a framework to manage the provenance and generation conditions of AIGC is essential. In this paper, we present a framework that automatically attaches structured metadata to AIGC at generation time, including modularized prompts, contexts, thoughts, model information, hyperparameters, and confidence. The metadata is enveloped together with verifiable credentials to support the reliable assessment and reuse of AIGC. This framework enables efficient curation of structured AIGC and facilitates its safe use for applications such as fine-tuning and knowledge distillation.

---

### TileQ: Efficient Low-Rank Quantization of Mixture-of-Experts with 2D Tiling

Authors: Hongyaoxing Gu, Xinzhe Chen, Lijuan Hu, Fangfang Liu
Published: 2026-05-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.09281v1

Mixture-of-Experts (MoE) models achieve remarkable performance by sparsely activating specialized experts, yet their massive parameters in experts pose significant challenges for deployment. While low-rank quantization offers a promising route to compress MoE models, existing methods still incur nonnegligible memory overhead and inference latency. To address these limitations, we propose \textsc{TileQ}, a fine-tuning-free post-training quantization (PTQ) method that employs 2D-tiling structured low-rank quantization to share low-rank factors across both input and output dimensions of MoE experts. Furthermore, we introduce an efficient inference technique for \textsc{TileQ} that fuses multiple low-rank expert computations into a single-pass operation, significantly improving hardware utilization. Experiments show that \textsc{TileQ} cuts down additional memory usage up to 10$\times$ and reduces inference latency to $\sim$5\% while preserving state-of-the-art accuracy.

---

### CAGS: Color-Adaptive Volumetric Video Streaming with Dynamic 3D Gaussian Splatting

Authors: Daheng Yin, Yili Jin, Jianxin Shi, Isaac Ding, Miao Zhang, Fangxin Wang, Zhaowu Huang, Cong Zhang, Jiangchuan Liu, Fang Dong
Published: 2026-05-10
Categories: cs.GR, cs.CV, cs.MM, cs.NI, eess.IV
arXiv: https://arxiv.org/abs/2605.09279v1

Volumetric video (VV) streaming enables real-time, immersive access to remote 3D environments, powering telepresence, ecological monitoring, and robotic teleoperation. These applications turn VV streaming into a real-time interface to remote physical environments, imposing new system-level demands for photorealistic scene representation, low-latency interaction, and robust performance under heterogeneous networks. 3D Gaussian Splatting (3DGS) has been widely used for real-time photorealistic rendering, offering superior visual quality and rendering performance, but it faces challenges due to bandwidth consumption. Furthermore, as the foundation of adaptive VV streaming, existing Levels of Detail (LoD) methods based on density are not well-suited to Gaussian representations, leading to visible gaps and severe quality degradation. Recent studies have also explored attribute compression techniques to reduce bandwidth consumption. Our preliminary studies reveal that aggressive attribute compression primarily causes color distortion, which can be effectively corrected in the rendered image using a reference image. Motivated by these findings, we propose a novel Color-Adaptive scheme for adaptive VV streaming that uses vector quantization (VQ) to establish LoDs and correct color distortions with low-resolution reference images. We further present CAGS, an adaptive VV streaming system compatible with diverse Gaussian representations, which integrates the Color-Adaptive scheme by rendering reference images on the streaming server and performing color restoration on the client. Extensive experiments on our prototype system demonstrate that CAGS outperforms the existing adaptive streaming systems in PSNR by 5$\sim$20 dB under fluctuating bandwidth, operates significantly faster than existing scalable Gaussian compression methods, and generalizes across different Gaussian representations.

---

### EquiMem: Calibrating Shared Memory in Multi-Agent Debate via Game-Theoretic Equilibrium

Authors: Yuqiao Meng, Luoxi Tang, Sakshi Sunil Narvekar, Rupali Rajendra Vaje, Yingxue Zhang, Muchao Ye, Zhaohan Xi
Published: 2026-05-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2605.09278v2

Multi-agent debate (MAD) systems increasingly rely on shared memory to support long-horizon reasoning, but this convenience opens a critical vulnerability: a single corrupted entry can contaminate the downstream memory-augmented reasoning, and debate alone fails to filter such errors. Existing safeguards filter entries via heuristics or LLM-based validation, yet they rely on AI judgments that share the same failure modes and overlook the cross-agent dynamics of MAD. We address this gap by formulating memory updating in MAD as a zero-trust memory game, in which no agent is assumed reliable and the game's equilibrium motivates a principled objective for calibrating memory influence. Guided by this objective, we propose EquiMem, an inference-time calibration mechanism that evaluates each update against the shared memory state and the memory already being used by other agents, using their existing retrieval queries and traversal paths as evidence of calibration. EquiMem instantiates this calibration for both embedding- and graph-based memory, and across diverse benchmarks, MAD frameworks, and memory architectures, it consistently outperforms existing safeguards, remains robust under adversarial agents, and incurs negligible inference overhead.

---

### EquiMem: Calibrating Shared Memory in Multi-Agent Debate via Game-Theoretic Equilibrium

Authors: Yuqiao Meng, Sakshi Sunil Narvekar, Luoxi Tang, Rupali Rajendra Vaje, Yingxue Zhang, Muchao Ye, Zhaohan Xi
Published: 2026-05-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2605.09278v1

Multi-agent debate (MAD) systems increasingly rely on shared memory to support long-horizon reasoning, but this convenience opens a critical vulnerability: a single corrupted entry can contaminate the downstream memory-augmented reasoning, and debate alone fails to filter such errors. Existing safeguards filter entries via heuristics or LLM-based validation, yet they rely on AI judgments that share the same failure modes and overlook the cross-agent dynamics of MAD. We address this gap by formulating memory updating in MAD as a zero-trust memory game, in which no agent is assumed honest and the game's equilibrium serves as an indicator of optimal memory trust. Guided by this equilibrium, we propose EquiMem, an inference-time calibration mechanism that quantifies each update algorithmically against the shared memory state, using agents' existing retrieval queries and traversal paths as evidence rather than soliciting any LLM judgment. EquiMem instantiates calibration for both embedding- and graph-based memory, and across diverse benchmarks, MAD frameworks, and memory architectures, it consistently outperforms existing safeguards, remains robust under adversarial agents, and incurs negligible inference overhead.

---

### Worst-Case Regret Bounds for Combinatorial Thompson Sampling in Sleeping Semi-Bandits

Authors: Zhiming Huang, Bingshan Hu, Jianping Pan
Published: 2026-05-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.09277v2

We revisit combinatorial Thompson sampling (CTS) for semi-bandits with sleeping arms, where arm availability varies over time and actions must satisfy combinatorial constraints, as in wireless mesh routing with fluctuating link availability. Despite its practical relevance, CTS has been hindered by several long-standing problems: (i) the absence of worst-case regret guarantees in the semi-bandit setting even without sleeping arms, (ii) the lack of theory under adversarially varying availability, and (iii) the consistently weak empirical performance of CTS with Gaussian priors (CTS-G). This paper resolves these long-standing issues by providing the first worst-case regret analysis of CTS-G, proving an upper bound of $\tilde{O}(m\sqrt{NT})$ and a matching lower bound of $\tildeΩ(m\sqrt{NT})$. To bridge the gap between theory and practice, we further propose CL-SG, a simple CTS-G variant that samples a single shared Gaussian seed each round to coordinate exploration across arms. We show that CL-SG achieves an improved regret bound of $\tilde{O}(\sqrt{mNT})$, together with a matching lower bound $Ω(\sqrt{mNT})$. Experiments on real-world datasets demonstrate that CL-SG consistently outperforms strong baselines including CTS-G and CTS-B, and we open-source our implementation for reproducibility.

---

### Uncertainty-Aware Token Importance Estimation in Spiking Transformers

Authors: Wenxuan Liu, Zecheng Hao, Tong Bu, Yuran Wang, Zhaofei Yu
Published: 2026-05-10
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2605.09276v1

Spiking transformers have shown strong potential for neuromorphic vision, yet their token processing across multiple spiking steps still introduces substantial redundancy and inference cost. Existing token reduction methods mainly rely on response based cues, such as activation magnitude, firing statistics, or feature similarity. Although effective, these criteria do not explicitly characterize token importance from the perspective of temporally evolving class evidence. In spiking transformers, token representations are progressively formed across multiple spiking steps rather than determined at a single instant, suggesting that token importance should be evaluated not only by instantaneous responses but also by temporal uncertainty patterns. Our key observation is that tokens exhibit heterogeneous uncertainty trajectories over time, and that their temporally aggregated uncertainty statistics provide an effective cue for distinguishing informative tokens from redundant ones. Motivated by this, we propose Uncert, a training free and plug and play token importance estimation framework for spiking transformers. Specifically, Uncert models token wise class evidence with a Dirichlet distribution and summarizes each token temporal uncertainty using its mean and fluctuation across spiking steps, yielding an uncertainty aware importance score for token reduction during inference. Experiments on both static and neuromorphic benchmarks show that Uncert achieves favorable accuracy and efficiency tradeoffs, with the most consistent gains observed under token pruning. Further analysis reveals a clear empirical connection between temporal uncertainty patterns and token contribution, offering new insights into token dynamics in spiking transformers.

---

### DiffATS: Diffusion in Aligned Tensor Space

Authors: Jinhua Lyu, Tianmin Yu, Brian Kim, Lizhuo Zhou, Chanwook Park, Naichen Shi
Published: 2026-05-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.09275v1

Direct diffusion modeling of high-resolution spatiotemporal fields is computationally challenging. Parameter-efficient primitives address this by representing high-dimensional data with a compact set of parameters. In this paper, we construct data-dependent tensor primitives without pretrained compression autoencoders. Our construction starts from Tucker decomposition, which captures low-rank multilinear structure through a core tensor and mode-wise factors. However, Tucker factors are non-unique: the same tensor can be represented by different rotated factors, which complicates generative modeling. We address this issue with orthogonal Procrustes (OP) alignment. Specifically, we select medoid anchor matrices from the data and align the factor matrices to resolve the gauge ambiguity. This yields matrix Grassmannian primitives and tensor Grassmannian primitives that are compact, data-adaptive, and directly decodable by explicit multilinear reconstruction. Theoretically, we prove that the proposed primitive maps are homeomorphisms between low-rank tensors and their corresponding primitive spaces, certifying that the representations are non-degenerate and topologically faithful. Building on these primitives, we propose *Diffusion in Aligned Tensor Space* (DiffATS), a generative framework that trains diffusion models directly on aligned tensor primitives. Across images, videos, and PDE solutions, DiffATS achieves strong unconditional and conditional generation performance while compressing original data by $3.9\times$ to $210\times$, without relying on any pretrained deep compression autoencoders.

---

### Instance-Adaptive Online Multicalibration

Authors: Zhiming Huang, Jamie Morgenstern, Aaron Roth, Claire Jie Zhang
Published: 2026-05-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.09273v3

We study online multicalibration beyond the worst-case. We give a single, efficient algorithm which dynamically interpolates between benign and worst-case sequences by adaptively refining a dyadic grid of prediction values. Its error is controlled by the number of leaves in the refinement tree. Our analysis recovers the known $\widetilde O(T^{2/3})$ worst-case-optimal rate for online multicalibration, while simultaneously automatically adapting to easier instances: in the marginal stochastic setting it obtains a rate of $\widetilde O(\sqrt T)$, and for piecewise-stationary means with $J$ segments its rate is $\widetilde O(\sqrt{JT})$. More generally, the rate depends on a threshold-complexity measure of the predictable mean process relative to the group family. We show that this dependence is tight up to logarithmic factors.

---

### Instance-Adaptive Online Multicalibration

Authors: Zhiming Huang, Jamie Morgenstern, Aaron Roth, Claire Jie Zhang
Published: 2026-05-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.09273v2

We study online multicalibration beyond the worst-case. We give a single, efficient algorithm which dynamically interpolates between benign and worst-case sequences by adaptively refining a dyadic grid of prediction values. Its error is controlled by the number of leaves in the refinement tree. Our analysis recovers the known $\widetilde O(T^{2/3})$ worst-case-optimal rate for online multicalibration, while simultaneously automatically adapting to easier instances: in the marginal stochastic setting it obtains a rate of $\widetilde O(\sqrt T)$, and for piecewise-stationary means with $J$ segments its rate is $\widetilde O(\sqrt{JT})$. More generally, the rate depends on a threshold-complexity measure of the predictable mean process relative to the group family. We show that this dependence is tight up to logarithmic factors.

---

### Towards Conversational Medical AI with Eyes, Ears and a Voice

Authors: Meet Shah, Jason Gusdorf, Anil Palepu, Chunjong Park, Jack W. O'Sullivan, Vishnu Ravi, Tim Strother, Pavel Dubov, Aliya Rysbek, Toshiyuki Fukuzawa, Yana Lunts, Jan Freyberg et al.
Published: 2026-05-10
Categories: cs.AI, cs.CL, cs.CV
arXiv: https://arxiv.org/abs/2605.09272v1

The practice of medicine relies not only upon skillful dialogue but also on the nuanced exchange and interpretation of rich auditory and visual cues between doctors and patients. Building on the low-latency voice and video processing capabilities of Gemini, we introduce AI co-clinician, a first-of-its-kind conversational AI system utilizing continuous streams of audio-visual data from live patient conversations to inform real-time clinical decisions. Its dual-agent architecture balances deep clinical reasoning with the low latency required for natural dialogue. To assess this system, we implemented a video-based interface emulating telemedicine consultations. We crafted 20 standardized outpatient scenarios requiring proactive real-time auditory and visual reasoning and designed "TelePACES" evaluation criteria alongside case-specific rubrics. In a randomized, interface-blinded, crossover simulation study (n = 120 encounters) with 10 internal medicine residents as patient actors, we compared AI co-clinician with primary care physicians (PCPs), GPT-Realtime, and a baseline agent. AI co-clinician approached PCPs in key TelePACES dimensions, including management plans and differential diagnosis, while significantly outperforming GPT-Realtime across all general criteria. While our agent demonstrated parity with PCPs in case-specific triage measures, physicians maintained superior overall performance in case-specific assessments. Although AI co-clinician marks a significant advance in real-time telemedical AI, gaps remain in physical examination and disease-specific reasoning. Our work shows that text-only approaches fail to capture the true challenges of medical consultation and suggests that high-stakes real-time diagnostic AI is most safely advanced in collaborative, triadic models where AI can be a supportive co-clinician for doctors and patients.

---

### Shaping Schema via Language Representation as the Next Frontier for LLM Intelligence Expanding

Authors: Zhiqin Yang, Yuhan Liu, Jingwen Fu, Pei Fu, Bo Han, Masashi Sugiyama, Nanning Zheng
Published: 2026-05-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2605.09271v1

Although natural language is the default medium for Large Language Models (LLMs), its limited expressive capacity creates a profound bottleneck for complex problem-solving. While recent advancements in AI have relied heavily on scaling, merely internalizing knowledge does not guarantee its effective application. Defining language representation as the linguistic and symbolic constructs used to map and model the real world, this paper argues that shaping schemas through advanced language representation is the next frontier for expanding LLM intelligence. We posit that an LLM's knowledge activation and organization -- its schema -- depends heavily on the structural and symbolic sophistication of the language used to represent a given task. This paper contributes both a formalization of this claim and the empirical evidence to support it. With a new formalization, we present multiple lines of evidence to support our position: Firstly, we review recent empirical practices and emerging methodologies that demonstrate the substantial performance gains achievable through deliberate language representation design, even without modifying model parameters or scale. Secondly, we conduct controlled experiments showing that LLM performance and its internal feature activations vary under different language representations of the same underlying task. Together, these findings highlight language representation design as a promising direction for future research.

---

### Memorize Theorems, Not Instances: Probing SFT Generalization through Mathematical Reasoning

Authors: Ruiying Peng, Mengyu Yang, Jing Lei, Xiaohui Li, Xueyu Wu, Xinlei Chen
Published: 2026-05-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2605.09270v2

Supervised Fine-Tuning (SFT) is widely used for task-specific adaptation, yet recent work shows it systematically undermines reasoning generalization. We argue the root cause is not memorization itself, but its target: vanilla SFT drives models to exploit and memorize spurious surface correlations in problem-solution pairs, leaving them brittle to superficial input variations. To address this, we propose Theorem-SFT, which reorients supervision toward explicit theorem application by teaching models how rules are invoked rather than what answers look like. Theorem-SFT yields consistent gains across benchmarks and model families: +8.8% on MATH (LLaMA3.2-3B-Instruct) and +20.27% on GeoQA (Qwen2.5-VL-7B-Instruct) without modality-specific re-training. Fine-tuning MLP layers alone matches full-layers performance, implicating feed-forward components as the primary locus of reasoning rules. Our findings reframe the debate: Generalization failures stem not from memorization as a mechanism, but from memorizing the wrong inductive targets.

---

### DeltaRubric: Generative Multimodal Reward Modeling via Joint Planning and Verification

Authors: Rui Liu, Dian Yu, Zhenwen Liang, Yucheng Shi, Tong Zheng, Runpeng Dai, Haitao Mi, Pratap Tokekar, Leoweiliang
Published: 2026-05-10
Categories: cs.CL, cs.CV
arXiv: https://arxiv.org/abs/2605.09269v1

Aligning Multimodal Large Language Models (MLLMs) requires reliable reward models, yet existing single-step evaluators can suffer from lazy judging, exploiting language priors over fine-grained visual verification. While rubric-based evaluation mitigates these biases in text-only settings, extending it to multimodal tasks is bottlenecked by the complexity of visual reasoning. The critical differences between responses often depend on instance-specific visual details. Robust evaluation requires dynamically synthesizing rubrics that isolate spatial and factual discrepancies. To address this, we introduce $\textbf{DeltaRubric}$, an approach that reformulates multimodal preference evaluation as a plan-and-execute process within a single MLLM. DeltaRubric operates in two steps: acting first as a $\textit{Disagreement Planner}$, the model generates a neutral, instance-specific verification checklist. Transitioning into a $\textit{Checklist Verifier}$, it executes these self-generated checks against the image and question to produce the final grounded judgment. We formulate DeltaRubric as a multi-role reinforcement learning problem, jointly optimizing planning and verification capabilities. Validated on Qwen3-VL 4B and 8B Instruct models, DeltaRubric achieves solid empirical gains. For instance, On VL-RewardBench, it improves base model overall accuracy by $\textbf{+22.6}$ (4B) and $\textbf{+18.8}$ (8B) points, largely outperforming standard no-rubric baselines. The results demonstrate that decomposing evaluation into structured, verifiable steps leads to more reliable and generalizable multimodal reward modeling.

---

### Beyond Continuity: Challenges of Context Switching in Multi-Turn Dialogue with LLMs

Authors: Aditya Sinha, Harald Steck, Vito Ostuni, Matteo Rinaldi
Published: 2026-05-10
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2605.09268v1

Users interacting with Large Language Models (LLMs) in a multi-turn conversation routinely refine their requests or pivot to new topics. LLMs, however, often miss these topic shifts and carry over irrelevant context from previous turns, leading to inaccurate responses. In this paper, we stress-test the multi-turn understanding of LLMs and study the following two sub-tasks: (1) detecting whether the user pivots or refines in the current turn, and (2) shortlisting relevant context from previous turns. To this end, we construct synthetic benchmarks based on real-world datasets from varied domains, as to simulate context shifts of different levels of difficulty. We then evaluate the zero-shot performance of ten LLMs (open-weight, closed-source and reasoning), and demonstrate that only some reasoning and strongly instructed LLMs are accurate in detecting pivots; open-weight LLMs struggle with the task and frequently carry stale context even with explicit cues; and all models suffer from a position bias. Based on the results, we discuss key takeaways for improving long-term robustness in multi-turn capabilities for LLMs.

---

### SeePhys Pro: Diagnosing Modality Transfer and Blind-Training Effects in Multimodal RLVR for Physics Reasoning

Authors: Kun Xiang, Terry Jingchen Zhang, Zirong Liu, Bokai Zhou, Yueling Tang, Junjie Yu, Jiacong Lu, Shangrui Huang, Heng Li, Likui Zhang, Kunkun Liu, Changzheng Zhang et al.
Published: 2026-05-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2605.09266v2

We introduce SeePhys Pro, a fine-grained modality transfer benchmark that studies whether models preserve the same reasoning capability when critical information is progressively transferred from text to image. Unlike standard vision-essential benchmarks that evaluate a single input form, SeePhys Pro features four semantically aligned variants for each problem with progressively increasing visual elements. Our evaluation shows that current frontier models are far from representation-invariant reasoners: performance degrades on average as information moves from language to diagrams, with visual variable grounding as the most critical bottleneck. Motivated by this inference-time fragility, we further develop large training corpora for multimodal RLVR and use blind training as a diagnostic control, finding that RL with all training images masked can still improve performance on unmasked validation sets. To analyze this effect, text-deletion, image-mask-rate, and format-saturation controls suggest that such gains can arise from residual textual and distributional cues rather than valid visual evidence. Our results highlight the need to evaluate multimodal reasoning not only by final-answer accuracy, but also by robustness under modality transfer and by diagnostics that test whether improvements rely on task-critical visual evidence.

---

### Reinforcing Multimodal Reasoning Against Visual Degradation

Authors: Rui Liu, Dian Yu, Haolin Liu, Yucheng Shi, Tong Zheng, Runpeng Dai, Haitao Mi, Pratap Tokekar, Leoweiliang
Published: 2026-05-10
Categories: cs.CV, cs.CL
arXiv: https://arxiv.org/abs/2605.09262v1

Reinforcement Learning has significantly advanced the reasoning capabilities of Multimodal Large Language Models (MLLMs), yet the resulting policies remain brittle against real-world visual degradations such as blur, compression artifacts, and low-resolution scans. Prior robustness techniques from vision and deep RL rely on static data augmentation or value-based regularization, neither of which transfers cleanly to critic-free RL fine-tuning of autoregressive MLLMs. Reinforcing reasoning against such corruptions is non-trivial: naively injecting degraded views during rollout induces reward poisoning, where perceptual occlusions trigger hallucinated trajectories and destabilize optimization. We propose ROMA, an RL fine-tuning framework that modifies the optimization dynamics to reinforce reasoning against visual degradation while preserving clean-input performance. A dual-forward-pass strategy uses teacher forcing to evaluate corrupted views against clean-image trajectories, avoiding new rollouts on degraded inputs. For distributional consistency, we apply a token-level surrogate KL penalty against the worst-case augmentation; to prevent policy collapse under regularization, an auxiliary policy gradient loss anchored to clean-image advantages preserves a reliable reward signal; and to avoid systematically incorrect invariance, correctness-conditioned regularization restricts enforcement to successful trajectories. On Qwen3-VL 4B/8B across seven multimodal reasoning benchmarks, our method improves robustness by +2.4% on seen and +2.3% on unseen corruptions over GRPO while matching clean accuracy.

---

### Remix the Timbre: Diffusion-Based Style Transfer Across Polyphonic Stems

Authors: Leduo Chen, Junchuan Zhao, Shengchen Li
Published: 2026-05-10
Categories: cs.SD, cs.AI
arXiv: https://arxiv.org/abs/2605.09259v1

Timbre transfer aims to modify the timbral identity of a musical recording while preserving the original melody and rhythm. While single-instrument timbre transfer has made substantial progress, existing approaches to multi-instrument settings rely on separate-then-transfer pipelines that propagate source separation artifacts and produce incoherent synthesized timbres across stems. This paper proposes MixtureTT, to the best of our knowledge the first system for flexible per-stem timbre transfer directly from a polyphonic mixture. Given a mixture and a separate timbre reference for each target voice, MixtureTT jointly transfers all stems to the specified instruments through a shared diffusion process. Modeling the dependencies across the per-stem content and cross-stem harmonic, the proposed joint stem diffusion transformer eliminates cascaded separation error, reduces inference cost by a factor equal to the number of stems, and yields more coherent multi-stem outputs. Despite operating under a strictly harder input condition, evaluations on the SATB choral dataset show that MixtureTT outperforms single-instrument baselines on both objective and subjective metrics demonstrating the necessity of dedicated multi-instrument timbre transfer over the naive separate-then-transfer pipelines. As a result, this work confirms that the cross-stem modeling is essential for mixture-level timbre transfer as the proposed joint setting consistently exceeds an equivalent single-stem ablation.

---

### Monocular Biomechanical Tracking of Fingers with Inverse Kinematics to Foundation Models

Authors: R. James Cotton, Pouyan Firouzabadi, Wendy Murray
Published: 2026-05-10
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2605.09258v1

Accurate hand and finger tracking from video has significant clinical applications for monitoring activities of daily living and measuring range of motion, yet monocular video approaches for obtaining hand biomechanics remain under-developed. We present a method that combines the SAM 3D Body foundation model with inverse kinematics optimization in a full-body biomechanical model to extract anatomically-constrained finger joint angles from single-view video. We port SAM 3D Body from PyTorch to JAX for integration with MuJoCo-MJX, enabling GPU-accelerated optimization, and develop a novel mapping between the Momentum Human Rig (MHR) outputs and biomechanical model markers. Validation against 8-camera multiview reconstruction on 4,590 frames from 7 participants performing a variety of hand poses and object manipulation tasks shows finger joint angle errors of approximately 10 degrees and hand position errors of approximately 6 mm, after Procrustes alignment. Results were consistent across camera viewpoints and robust to different methods for producing reference values from multiview video. This work extends monocular biomechanical analysis to detailed finger tracking, expanding access to quantitative characterization of hand movement from readily available video.

---

### Improving Generalization by Permutation Routing Across Model Copies

Authors: Shuhei Kashiwamura, Timothee Leleu
Published: 2026-05-10
Categories: cs.LG, cs.AI, stat.ML
arXiv: https://arxiv.org/abs/2605.09256v1

We introduce a use of the \(M\)-cover (or \(M\)-layer) transform for machine learning. The method replicates a model \(M\) times, but instead of coupling the copies through parameter averaging or an explicit attractive force, as in replicated SGD or Elastic SGD, it rewires the contexts in which local learning messages are computed. Each local loss is evaluated on a routed model whose parameters are drawn from different copies according to permutations sampled from a structured mixing kernel \(Q\). Training then uses the original local update rule, while the resulting learning messages are redistributed across the copies through these routed computational paths. Thus \(Q\) defines a topology for message transport and controls the long-loop structure of the lifted factor graph. We formulate this construction for perceptrons, committee machines, and multilayer perceptrons, showing that the same principle applies from discrete models to differentiable neural networks. The resulting framework provides a mechanism for improving generalization through structured message sharing rather than replica collapse or parameter-space coupling.

---