Explore the Frontier of Global Academia

01.

arXiv (math.PR) 2026-06-11 DOI: arXiv:2606.11902

Delta-Epsilon-Common Knowledge and Quantitative Agreement Theorems

Authors:

Christina Pawlowitsch↗Stefan Schrott↗Daniel Toneian↗

arXiv:2606.11902v1 Announce Type: cross Abstract: Aumann defined common knowledge mathematically and established his now famous Agreement Theorem. We present a novel approach to quantifying how close individuals are to commonly knowing events, $(\delta,\epsilon)$-common knowledge, which is defined for any (and not just countable) probability spaces, and provide quantitative versions of the key results in this field. Specifically, we do this for Aumann's Agreement Theorem and Nielsen's extension thereof to random variables, as well as for the setting in which posteriors are communicated back and forth between individuals. Our results apply in particular to noisy communication settings.

Read & Discuss → View Source →

02.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2606.16617

Sycophancy as Material Failure under Pushback Loading: A Multi-Axis Characterization Across Three Loading Cases and up to Seventeen Material Charges

Authors:

Ferdinand M. Schessl↗

Sycophancy in LLMs is documented across 70+ papers, but expert agreement on construct boundaries remains low (ICC=.184; Ye et al., 2026). The construct fragments because behavioral classification depends on which surface form is privileged. We adopt a materials-science framing: conversation as test specimen under load, LLM-model as material charge, pushback as progressive load, stance-flip as material failure. We characterize this failure across three loading cases (debate n=1000; false-presuppositions n=3400; ethical-setting n=3400; 10-17 material charges per case; 7800 specimens total) using 14 turn-level axis-measurements spanning velocity, damage accumulation, frame-drift, brittleness, and direction stability, plus three speaker-resolved axes from an independent pipeline. The measurements are Hooke-coupled ($\sigma = E \cdot \varepsilon$ analog) and reproduce across loading cases with effects up to $|r_{rb}| = 0.35$ on debate; the sign structure adds a second pattern: the ethical-setting case inverts the velocity and accumulation blocks. Variance composition partitions into two profiles: debate is charge-dominated (brittle-fracture-like: the material grade decides), false-presuppositions and ethical-setting are topic-dominated (creep-like: the load decides); the ratios (2.03 vs 0.13/0.17) are estimator-dependent, for debate even in direction. Cross-judge reliability (GPT-4o vs Haiku 4.5) shows debate scoring is judge-robust (Cohen's $\kappa = 0.88$) while false-presupposition scoring is judge-sensitive ($\kappa = 0.36$) – a caveat single-judge benchmarks must report. This is the methodological move Ye et al.'s diagnosis calls for: a multi-axis characterization that does not depend on which surface form of the construct one privileges.

Read & Discuss → View Source →

03.

arXiv (CS.CV) 2026-06-12 DOI: arXiv:2606.12886

Bridging Modal Isolation in Interleaved Thinking: Supervising Modality Transitions via Stepwise Reinforcement

Authors:

Tingyu Li↗Le Zhou↗Siyuan Li↗Yujun Wu↗Xinglong Xu↗Jingxuan Wei↗Conghui He↗Cheng Tan↗

Interleaved thinking, where a unified multimodal model alternates between textual reasoning and visual generation, has shown promise on spatial and physical tasks. However, in complex long-chain scenarios, we identify a fundamental failure mode: generated images diverge from the textual context while subsequent text ignores the visual evidence, causing the two modalities to alternate without genuinely informing each other. We term this Modal Isolation and attribute it to compounding information loss at modality boundaries. We decompose each reasoning cycle into atomic operations and define modality transition loss, quantifying cross-modal hallucination (text-to-image) and visual utilization deficit (image-to-text) at each boundary. We propose MoTiF (Modality Tiransition Fidelity), a two-stage training framework that directly optimizes these transitions: Reflective SFT trains the model to detect and recover from erroneous visual outputs; Flow-GRPO improves image generation fidelity via reinforcement learning. All training signals in MoTiF derive from transition-level fidelity rather than end-task accuracy. Across four visual puzzle benchmarks, this transition-level supervision substantially improves both cross-modal coherence and final task accuracy. The results demonstrate that effective interleaved reasoning requires explicit structural supervision at modality boundaries, not merely scaling or end-task optimization.

Read & Discuss → View Source →

04.

arXiv (CS.CV) 2026-06-17 DOI: arXiv:2606.17480

GeneralVLA-2: Geometry-Aware Reconstruction and Governed Memory for Robot Planning

Authors:

Haoyu Wang↗Guoqing Ma↗Zeyu Zhang↗Yandong Guo↗Boxin Shi↗Hao Tang↗

Generalist vision-language-action systems need object-centric 3D evidence and reusable manipulation experience to plan reliable robot trajectories. GeneralVLA provides a hierarchical interface for converting language and RGB-D observations into 3D end-effector paths, but two bottlenecks remain. First, monocular SAM3D-style object reconstruction can hallucinate pose and unseen geometry, while manipulation benefits from stable object shape when calibrated multi-view observations are available. Second, the original KnowledgeBank mainly retrieves semantically similar snippets and appends new knowledge, which makes it difficult to control memory quality, conflicts, confidence, and geometric relevance. To address the first challenge, we introduce GeoFuse-MV3D, a geometry-prior-guided MV-SAM3D reconstruction branch that verifies external geometry cues with input-view masks, applies soft visual-hull support, performs axis-wise refinement, and fuses only geometry while preserving appearance. To address the second challenge, we upgrade KnowledgeBank into a governed long-term memory system with explicit quality, confidence, lifecycle, verifier, and conflict metadata, together with precision-oriented retrieval. Finally, we evaluate the reconstruction branch on GSO-30 and the memory module on Terminal-Bench 2.0 and SWE-Bench Verified; GeoFuse-MV3D improves over the MV-SAM3D baseline by reducing CD and LPIPS by 2.20% and 2.02% while increasing PSNR and SSIM by 2.36% and 1.03%, and KnowledgeBank improves over ReasoningBank by 4.53% on Terminal-Bench SR and 3.73% on SWE-Bench resolve rate, while reducing AS by 4.95% and 5.65%, respectively. Code: https://github.com/AIGeeksGroup/GeneralVLA-2. Website: https://aigeeksgroup.github.io/GeneralVLA-2.

Read & Discuss → View Source →

05.

arXiv (CS.CL) 2026-06-19 DOI: arXiv:2604.23938

TSAssistant: A Human-in-the-Loop Agentic Framework for Automated Target Safety Assessment

Authors:

Xiaochen Zheng↗Zhiwen Jiang↗David Tokar↗Yexiang Cheng↗Alvaro Serra↗Melanie Guerard↗Klas Hatje↗Tatyana Doktorova↗

Target Safety Assessment (TSA) requires systematic integration of genetic, transcriptomic, target homology, pharmacological, and clinical data to evaluate potential safety liabilities of therapeutic targets. This process is labor-intensive and expert-dependent, posing challenges in scalability and reproducibility. We present TSAssistant, a human-in-the-loop multi-agent framework that decomposes TSA report generation into a workflow of specialized subagents: Research Subagents that each ground and cite a single TSA domain, and Synthesis Subagents that integrate findings across domains. Subagents retrieve and synthesize evidence from curated biomedical sources through standardized tool interfaces and produce individually citable, evidence-grounded sections, with behavior shaped by a hierarchical instruction architecture that separates coordination logic from domain expertise and user intent. To complement these soft constraints, programmatic execution hooks and persistent memory stores enforce hard constraints across the workflow, while an interactive refinement loop allows experts to review and revise individual sections with full conversational context preserved across iterations. Rather than a single holistic comparison, we decompose report quality into reproducibility, evidential grounding, task-level accuracy, and controllability under expert oversight, finding high reproducibility and grounding, substantial agreement with the human reference, and net-positive expert-driven refinement.

Read & Discuss → View Source →

06.

arXiv (CS.CV) 2026-06-18 DOI: arXiv:2606.18765

SpectralDiT: Timestep-Conditioned Spectral Residual Correction for Flow-Matching DiTs

Authors:

Jiayu Tian↗

We propose SpectralDiT, a lightweight modification to flow-matching Diffusion Transformers that adds timestep-conditioned spectral correction to the MLP residual branch. The module decomposes each residual update into low- and high-frequency components on the patch-token grid, then learns a zero-initialized additive gate so the model initially matches the baseline DiT. On CIFAR-10 pixel-space generation, SpectralDiT improves FID from 20.78 to 19.71 at patch size 1 and reduces the radial Fourier spectrum gap. Furthermore, we scale our method to latent diffusion on ImageNet-100. With 0.6% additional theoretical FLOPs and 1.36% additional parameters, SpectralDiT improves latent flow-matching, achieving an 8.7% relative FID reduction under classifier-free guidance (CFG 2.0). All reported results are averaged over five seeds. Ablations and gate visualizations on CIFAR-10 reveal stable block-specific spectral correction patterns.

Read & Discuss → View Source →

07.

arXiv (CS.LG) 2026-06-11 DOI: arXiv:2606.11891

Critic Architecture Matters: Dual vs. Unified Critics for Humanoid Loco-Manipulation

Authors:

Mehmet Turan Yard{\i}mc{\i}↗

arXiv:2606.11891v1 Announce Type: cross Abstract: Multi-objective reinforcement learning for humanoid robots must coordinate locomotion and manipulation within a single policy. A natural design choice is whether to use a single (unified) critic that estimates the combined value of all objectives, or separate (dual) critics with disjoint reward signals. We present a controlled comparison on the Unitree G1 humanoid (23 active DoF) in NVIDIA Isaac Lab, training loco-manipulation policies through a sequential curriculum spanning 13 levels from stationary reaching to walking with variable-orientation targets. In standardized evaluation, dual-critic policies reach targets 3.5$\times$ faster (6.5 vs. 22.6 simulation steps), achieve 2$\times$ higher throughput (14.3 vs. 7.0 validated reaches per 1,000 steps), and attain higher validated reach rates (65.2% vs. 53.8%) compared to the unified-critic policy. Notably, additional anti-gaming reward mechanisms provide no further improvement beyond the architectural change alone (60.9% vs. 65.2%). These results have direct implications for the emerging paradigm of RL fine-tuning of imitation-learned policies: when refining a pre-trained manipulation policy with RL, a unified critic risks suppressing the learned behavior through competing locomotion gradients. These findings demonstrate that critic architecture is a primary - and often overlooked - design choice in multi-objective humanoid RL, with greater impact than reward engineering on reaching efficiency.

Read & Discuss → View Source →

08.

arXiv (quant-ph) 2026-06-16 DOI: arXiv:2505.00588

Exact Many-body Quantum Dynamics in One-Dimensional Baths via Collective Spins

Authors:

Joseph T. Lee↗Silvia Cardenas-Lopez↗Stuart J. Masson↗Rahul Trivedi↗Ana Asenjo-Garcia↗

arXiv:2505.00588v2 Announce Type: replace Abstract: Computing the exact dynamics of many-body quantum systems becomes intractable as system size grows. Here, we present a symmetry-based method that provides an exponential reduction in the complexity of a broad class of such problems $\unicode{x2014}$ qubits coupled to one-dimensional electromagnetic baths. We identify conditions under which partial permutational symmetry emerges and exploit it to group qubits into collective multi-level degrees of freedom, which we term ''superspins.'' These superspins obey a generalized angular momentum algebra, reducing the relevant Hilbert space dimension from exponential to polynomial. Using this framework, we efficiently compute many-body superradiant dynamics in large arrays of qubits coupled to waveguides and ring resonators, showing that $\unicode{x2014}$ unlike in conventional Dicke superradiance $\unicode{x2014}$ the total spin length is not conserved. At long times, dark states become populated. We identify configurations where these states exhibit metrologically useful entanglement. Our approach enables exact treatment of complex dissipative dynamics beyond the fully symmetric limit and provides a rigorous benchmark for approximate numerical methods.

Read & Discuss → View Source →

09.

bioRxiv (Bioinfo) 2026-06-20 DOI: HASH:2291c7631b4dce607c6fd937b22d42ec

Seed variation impacts clustering stability in Single-Cell RNA-Seq and can be mitigated by StAbility-BasEd-Reassignment (SABER)

Authors:

Zagaria↗Tini↗Bonetti↗Mazzarella↗

Single-cell RNA-seq clustering is commonly treated as reproducible once a random seed is fixed, yet the choice of seed itself may alter cell assignments and downstream interpretation. We systematically quantified seed-induced clustering variability by running Louvain and Leiden clustering across 100 seeds in Seurat and Scanpy on 28 single-cell RNA-seq datasets from the Human Cell Atlas and IMMUcan. Using Element-Centric Consistency, we found that seed choice affected a substantial fraction of cells, with Scanpy showing more unstable assignments than Seurat on average, 40.46% versus 26.78% unstable cells, respectively. This increased stability came at a marked computational cost: Seurat required approximately 19-fold higher median memory than Scanpy. Seed-dependent clustering variability also propagated to cell-type annotation, particularly among transcriptionally related populations including macrophage/monocyte, endothelial/epithelial and T/NK cell states. To mitigate this instability, we developed StAbility-BasEd Reassignment (SABER), a Scanpy-based framework that identifies seed-sensitive cells across repeated clusterings and reassigns them to stable cluster cores using cosine similarity. SABER improved clustering quality while preserving annotation concordance and reduced median memory usage 3.5-fold compared with Seurat-Louvain. Our results identify seed choice as an underappreciated source of variability in single-cell analysis and provide a scalable strategy to improve clustering robustness.

Read & Discuss → View Source →

10.

arXiv (CS.CV) 2026-06-11 DOI: arXiv:2606.12218

Adapting Prithvi-EO for Fallow Detection for Food-Water Nexus: ViT-Adapter Necks and Parameter-Efficient Backbone tuning of Geospatial Foundation Model

Authors:

Sk Muhammad Asif↗Orhun Aydin↗

Understanding spatial distribution of fallow land is important for optimizing the food-water (FW) nexus, given fallowing's role in crop rotation and water conservation. Fallow is a low accuracy class in USDA Cropland Data Layer (CDL). Geospatial foundation model (GFM), Prithvi-EO has shown strong transferability across computer vision tasks. However, its Vision Transformer (ViT) backbone produces features at a single spatial scale that are ill-suited for the multi-scale features required by object detection heads. Existing approaches synthesise multi-scale pyramids through scaling of single stride tokens, sacrificing spatial heterogeneity, and full backbone fine-tuning is computationally prohibitive for GFMs. We evaluate a fallow detection pipeline combining two parameter-efficient fine tuning (PEFT) schemes: Low-Rank Adaptation (LoRA) and a hybrid PEFT, with three neck designs: pseudo multi-scale, Lite ViT-Adapter, and Full ViT-Adapter. Our best configuration, Lite ViT-Adapter with a one-stage head, achieves a mAP@50 of 0.9479 with the Diou loss, suggesting the effectiveness of center-aware localization for irregular fallow field detection. ViT-Adapter free one-stage detection under LoRA improves the adapter-free anchor-based approach by 6.42%, and the best configuration improves baseline adapter-free anchor-based approach by 25.70%. These results demonstrate that lightweight spatial prior fusion and selective backbone unfreezing enable Prithvi-EO to capture local fallow patterns more effectively, outperforming approaches that rely on reshaped single-stride ViT tokens.

Read & Discuss → View Source →

11.

arXiv (CS.CL) 2026-06-11 DOI: arXiv:2606.11643

Improving Cross-Format Robustness in Language Models with Multi-Format Training

Authors:

June M. Liu↗Shaomian Zheng↗He Cao↗Dingnan Jin↗Qing Cui↗Jun Zhou↗

Large language models often remain sensitive to answer format: a question solved correctly in one form may fail in another semantically equivalent form. To study this gap, we define cross-format robustness as the extent to which a model answers the same underlying question consistently across formats. We then compare full-format training with FormatMix, which expands only a subset of training items into multiple equivalent formats using either random or targeted selection. Across GLM4 and Llama-3.1, multi-format supervision consistently improves both task performance and cross-format robustness, whereas Multiple-choice question (MCQ)-only supervision alone brings little benefit and can even reduce robustness. We further find that expanding only about 30% of the training set into multiple formats often recovers most of the gain from full-format training, and this effect appears across the model families and sizes we study. These results suggest that format diversity, rather than additional supervision alone, is the key driver of robustness. That lightweight multi-format augmentation is a practical way to make LLMs less sensitive to answer format without changing the base model.

Read & Discuss → View Source →

12.

arXiv (CS.LG) 2026-06-11 DOI: arXiv:2606.12120

A Riemannian Approach to Low-Rank Optimal Transport

Authors:

Pratik Jawanpuria↗Bamdev Mishra↗

arXiv:2606.12120v1 Announce Type: new Abstract: Low-rank optimal transport (OT) mitigates the quadratic scaling of classical solvers, yet existing approaches rely heavily on first-order mirror-descent updates that require careful hyperparameter tuning and ignore the optimization landscape's curvature. To address these limitations, we propose a unified Riemannian geometric framework for low-rank OT, modeling balanced and unbalanced rank-$r$ positive factored couplings as novel smooth embedded submanifolds of the positive orthant. By equipping these manifolds with the Fisher-Rao product metric, we derive tractable formulations for Riemannian projectors, retractions, and Hessian-vector products. Our cost-agnostic framework seamlessly extends to linear OT, Gromov-Wasserstein (GW), fused GW, and their unbalanced counterparts. For balanced OT, our geometric ingredients are computed via efficient conjugate-gradient and iterative Bregman updates. For the unbalanced OT, our operations elegantly reduce to closed-form scalings, completely eliminating inner iterative loops. In both regimes, per-iteration complexity scales linearly with dataset size, and we provide a rank-sufficiency certificate for global optimality verification. Extensive experiments across a range of problem sizes demonstrate that our regularization-free first- and second-order solvers achieve faster convergence and superior performance over existing state-of-the-art low-rank OT solvers.

Read & Discuss → View Source →

13.

PLOS Medicine 2026-05-08 DOI: HASH:19df10f086a7c3b0f24990056e0dce8f

Optimal minimal residual disease threshold in pediatric acute myeloid leukemia: A retrospective cohort study based on the TARGET database

Authors:

Xiong-yu Liao↗

by Xiong-yu Liao, Hong Zheng, Jian-pei Fang, Dun-hua Zhou, Kun-yin Qiu Background Minimal residual disease (MRD) monitoring is a cornerstone of risk stratification in pediatric acute myeloid leukemia (AML), with a threshold of 0.1% conventionally defining positivity by flow cytometry. Advances in flow cytometric technologies, enabling detection of leukemic cells with higher sensitivity and specificity, warrant a reevaluation of whether a lower threshold improves prognostic accuracy. Methods and findings We conducted a retrospective cohort study using data from the Therapeutically Applicable Research to Generate Effective Treatments (TARGET)-AML initiative. The study population comprised 1,205 pediatric patients with de novo AML treated across Children’s Oncology Group (COG) clinical trial centers. Patients were enrolled between September 1996 and December 2016, with a median follow-up of 6.2 years (range: 0.5–20.1 years). The primary objective was to compare the prognostic performance of the traditional MRD threshold (≥0.1%) with a lower threshold (≥0.05%) after induction courses 1 and 2. The main outcome measure was 5-year event-free survival (EFS). Analyses included Kaplan−Meier survival estimates, Cox proportional hazards models to calculate hazard ratios (HR) with 95% confidence intervals (CI), receiver operating characteristic (ROC) curves, and net reclassification improvement (NRI). The optimal threshold for predicting 5-year EFS, determined by ROC analysis, was 0.05% after both induction course 1 (AUC: 0.840, 95%CI[0.76,0.88]) and course 2 (AUC: 0.854, 95%CI[0.78,0.89]). The 0.05% threshold demonstrated higher HR for the first event than the 0.1% threshold (after course 1: HR = 2.8, 95%CI[2.3,3.3]; P 

Read & Discuss → View Source →

14.

medRxiv (Medicine) 2026-06-18 DOI: HASH:3b55d63d01b63411f7058cb234476b5b

AlphaGenome identifies a deep intronic variant in a family with PLA2G6-associated neurodegeneration: Closing the diagnostic gap in rare genetic diseases

Authors:

Eger↗S. J↗Lopez↗Gomez Navarro↗L. F↗Pena-Tauber↗Cochran↗J. N↗Hiatt↗S. M↗Gelvez↗Garcia-Garcia↗…

A molecular diagnosis remains out of reach for a substantial subset of patients with clinically recognizable Mendelian disorders, even after comprehensive next-generation sequencing. Causal variants in non-coding regions are difficult to detect and interpret using standard pipelines. Deep intronic variants that disrupt splicing are a known but underexplored source of pathogenic alleles, and systematic tools to evaluate them at scale have only recently emerged. We aimed to resolve an incomplete genetic diagnosis in two siblings with early-onset parkinsonism, prominent neuropsychiatric features, and autonomic dysfunction consistent with PLA2G6-associated neurodegeneration (PLAN), an autosomal recessive condition. Prior clinical exome sequencing, genome sequencing, Multiplex Ligation-dependent Probe Amplification (MLPA), and long-read sequencing had identified only a single heterozygous PLA2G6 missense variant, c.2132C>G (p.Pro711Arg). We used AlphaGenome to score 91 non-coding variants shared among the affected siblings and their father within 1 megabase of the PLA2G6 locus. The deep-learning model identified an intronic variant (c.2034+355G>A) that was predicted to create a cryptic splice acceptor site that could result in inclusion of a 160-bp cryptic exon. Tissue-specific predictions indicated the aberrant splicing would be detectable in blood, confirmed by junction-spanning RNA-seq reads from an unrelated carrier. This analysis completed a compound heterozygous PLAN diagnosis nearly two decades after symptom onset and demonstrates the utility of sequence-to-function models. Systematic integration of tools like AlphaGenome into rare disease workflows offers a practical, low-barrier route to closing the diagnostic gap for patients with compelling Mendelian phenotypes and incomplete genetic diagnoses.

Read & Discuss → View Source →

15.

arXiv (CS.AI) 2026-06-19 DOI: arXiv:2507.19653

On the Limitations of Ray-Tracing for Learning-Based RF Tasks in Urban Environments

Authors:

Armen Manukyan↗Hrant Khachatrian↗Edvard Ghukasyan↗Theofanis P. Raptis↗

arXiv:2507.19653v2 Announce Type: replace-cross Abstract: We study the realism of Sionna v1.0.2 ray-tracing for outdoor cellular links in central Rome. We use a real measurement set of 1,664 user-equipments (UEs) and six nominal base-station (BS) sites. Using these fixed positions we systematically vary the main simulation parameters, including path depth, diffuse/specular/refraction flags, carrier frequency, as well as antenna's properties like its altitude, radiation pattern, and orientation. Simulator fidelity is scored for each base station via Spearman correlation between measured and simulated powers, and by a fingerprint-based k-nearest-neighbor localization algorithm using RSSI-based fingerprints. Across all experiments, solver hyper-parameters are having immaterial effect on the chosen metrics. On the contrary, antenna locations and orientations prove decisive. By simple greedy optimization we improve the Spearman correlation by 5% to 130% for various base stations, while kNN-based localization error using only simulated data as reference points is decreased by one-third on real-world samples, while staying twice higher than the error with purely real data. Precise geometry and credible antenna models are therefore necessary but not sufficient; faithfully capturing the residual urban noise remains an open challenge for transferable, high-fidelity outdoor RF simulation.

Read & Discuss → View Source →

16.

arXiv (CS.LG) 2026-06-12 DOI: arXiv:2606.12507

Rubric-Guided Self-Distillation: Post-Training Without Rubric Verifiers

Authors:

MohammadHossein Rezaei↗Anas Mahmoud↗Zihao Wang↗Utkarsh Tyagi↗Advait Gosai↗Razvan-Gabriel Dumitru↗Aakash Sabharwal↗Bing Liu↗Yunzhong He↗

arXiv:2606.12507v1 Announce Type: new Abstract: Rubrics have emerged as an alternative to RLVR in open-ended domains where a single ground-truth final answer is not available. Existing rubric-based training methods rely on an LLM verifier that scores each rollout against rubrics. This introduces substantial training-time overhead, exposes optimization to verifier-specific biases, and reduces rubric feedback to a sparse end-of-trajectory signal. We propose Rubric-Guided Self-Distillation (RGSD), a verifier-free training method in which the base policy, conditioned on the rubric, serves as the teacher for the unconditioned student. RGSD distills the rubric-conditioned teacher distribution into the student token-by-token, replacing sparse trajectory-level rewards with dense per-token learning signals and removing the LLM judge from the training loop entirely. Across Qwen-2.5 (3B, 7B) and Qwen3-Thinking (4B, 8B) models on medical and science domains, RGSD achieves rubric satisfaction comparable to judge-based GRPO while using one on-policy rollout per prompt and no training-time verifier calls. Ablations show that raw rubrics provide a stronger teacher enrichment signal than self-generated reference responses, while a stronger GRPO judge can outperform RGSD in some settings, positioning RGSD as a complementary verifier-free alternative when verifier cost or reliability is the bottleneck.

Read & Discuss → View Source →

17.

arXiv (CS.CL) 2026-06-19 DOI: arXiv:2606.20097

HydraHead: From Head-Level Functional Heterogeneity to Specialized Attention Hybridization

Authors:

Zhentao Tan↗Wei Chen↗Jingyi Shen↗Yao Liu↗Xu Shen↗Yue Wu↗Jieping Ye↗

The quadratic complexity of attention poses a critical bottleneck for long-context processing, spurring interest in hybrid attention designs. Most open-source hybrid models adopt a layer-wise strategy. Yet, prior work has noted the inherent difficulty of integrating Linear Attention (LA) with Full Attention (FA), suggesting that the design space of attention hybridization remains underexplored. To probe this space, we conduct interpretability analysis and observe that layers exhibit block-wise functional similarity, while individual heads within the same layer display distinct functional specialization despite sharing input features. This head-level heterogeneity suggests that the head dimension provides a natural and principled granularity for fusing heterogeneous attention signals. Building on this insight, we introduce HydraHead, a novel architecture that hybridizes FA and LA along the head axis. HydraHead features two key innovations: (1) an interpretability-driven selection strategy that identifies retrieval-critical heads and preserves FA only for them, and (2) a scale-normalized fusion module that reconciles the distributional gap between FA and LA head outputs. By leveraging a three-stage transfer pipeline with parameter reuse and distillation, we achieve high-performance hybrid models with minimal training overhead. Under a unified training setup, HydraHead outperforms other hybrid designs in long-context tasks while maintaining strong general reasoning. With interpretability-driven head selection, it matches a 3:1 layer-wise hybrid's long-context performance at a 7:1 LA-to-FA ratio. Crucially, trained on only 15B tokens, HydraHead achieves over 69% improvement over the baseline at 512K context length, approaching Qwen3.5, a leading model of comparable size with a native context length of 256K. This highlights the significant scaling potential of head-level hybridization.

Read & Discuss → View Source →

18.

arXiv (CS.CL) 2026-06-15 DOI: arXiv:2505.23277

Sentinel: Decoding Context Utilization via Attention Probing for Efficient LLM Context Compression

Authors:

Yong Zhang↗Heng Li↗Yanwen Huang↗Ning Cheng↗Yang Guo↗Yun Zhu↗Yanmeng Wang↗Shaojun Wang↗Jing Xiao↗

Retrieval-augmented generation (RAG) often suffers from long and noisy retrieved contexts. Existing context compression methods typically rely on heuristic relevance estimation or supervised compression models rather than on how LLMs utilize retrieved context during inference. We propose Sentinel, a lightweight sentence-level compression framework that decodes inference-time contextual utilization behaviors from head-wise attention patterns of frozen LLMs. To ground supervision in retrieval-dependent answering behavior, Sentinel trains a lightweight probe using QA examples where the model succeeds only when retrieved context is available. Sentinel performs compression using only a single non-autoregressive forward pass without dedicated compression training or autoregressive scoring. Empirically, we find that effective contextual utilization signals remain accessible even in compact proxy models. On LongBench, Sentinel with a 0.5B proxy model achieves up to 5$\times$ compression while attaining question-answering performance competitive with compression methods built on 7B-scale models. Despite being trained only on English QA data, Sentinel also generalizes effectively to Chinese and out-of-domain settings.

Read & Discuss → View Source →

19.

arXiv (CS.AI) 2026-06-18 DOI: arXiv:2606.19004

Spotlight: Synergizing Seed Exploration and Spot GPUs for DiT RL Post-Training

Authors:

Ruiqi Lai↗Dakai An↗Wei Gao↗Ju Huang↗Siran Yang↗Jiamang Wang↗Lin Qu↗Dmitrii Ustiugov↗Wei Wang↗

arXiv:2606.19004v1 Announce Type: cross Abstract: Reinforcement learning (RL) post-training of Diffusion Transformers (DiTs) is prohibitively expensive, requiring thousands of high-end GPUs. Existing works explore two directions to reduce cost: seed exploration improves training convergence by selecting high-contrast samples, yet adds compute to the critical path; spot GPUs offer 69–77\% lower cost, yet sit idle during training because DiT rollouts finish nearly simultaneously, which prevents LLM-style pipelining of rollout with training. Spot preemptions further break Sequence Parallelism (SP) groups, fragmenting GPU topology. We present Spotlight, the first system that harvests spot GPUs for DiT RL post-training. Spotlight rests on two key insights we devise: (1)~we show that exploration can tolerate stale model weights because exploration that uses the model weights from the previous iteration preserves the relative ranking of random seeds, allowing exploration to run on idle spot GPUs during training. (2)~SP reconfiguration can reuse on-node state, reducing group recovery from minutes to sub-second launches. Built on these insights, Spotlight introduces three techniques: a bandit-based exploration planner that maximizes reward variance within the training time budget, elastic sequence parallelism that reconfigures SP groups on the fly via persistent schedulers and intra-node weight copying, and a preemption-aware pull-based request scheduler that balances load and commits in-flight state upon preemption. We implement Spotlight on the open-source RL platform ROLL and evaluate it on Qwen-Image post-training. Spotlight reaches the same target validation score $4\times$ faster than baselines, reducing total cost by $1.4$-$6.4\times$ while achieving superior image quality on DeepSeek-OCR and Geneval datasets with resolution $512\times512$ and $1280\times1280$.

Read & Discuss → View Source →

20.

arXiv (CS.CL) 2026-06-11 DOI: arXiv:2606.12373

Verifiable Environments Are LEGO Bricks: Recursive Composition for Reasoning Generalization

Authors:

Hao Xiang↗Qiaoyu Tang↗Le Yu↗Yaojie Lu↗Xianpei Han↗Ben He↗Le Sun↗Bowen Yu↗Peng Wang↗Hongyu Lin↗Dayiheng Liu↗

Reinforcement Learning (RL) with verifiable environments has emerged as a powerful approach for enhancing the reasoning capabilities of Large Language Models (LLMs). While prior research demonstrates that scaling environment quantity improves RL performance, existing manual or individual construction methods suffer from linear scaling limits, thereby hindering scalable reasoning generalization. This paper introduces RACES (Recursive Automated Composition for Environment Scaling), a framework that conceptualizes verifiable environments as composable building blocks that can be recursively assembled. The key insight is that when the codomain (output type) of one environment matches the domain (input type) of another, they can be automatically fused into a new verifiable environment, enabling recursive composition. RACES is implemented with 300 individual environments and defines a set of composition operators (\textsc{SEQUENTIAL}, \textsc{PARALLEL}, \textsc{SORT}, and \textsc{SELECT}) that induce diverse reasoning patterns. Extensive experiments show that RL training on these composite environments consistently enhances reasoning generalization. Specifically, RACES improves DeepSeek-R1-Distill-Qwen-14B by an average of 3.1 points (from 48.2 to 51.3) and boosts Qwen3-14B performance from 58.8 to 61.1 on six benchmarks, which are unseen during the construction of training environments. Moreover, RACES achieves performance comparable to training on 300 individual environments using only 50 base environments, demonstrating significant efficiency in environment utilization.

Read & Discuss → View Source →

21.

arXiv (CS.AI) 2026-06-17 DOI: arXiv:2606.17847

WallZero: Mastering the Game of WallGo with Strategic Analysis

Authors:

Hsing-Yu Chen↗J\'er\^ome Arjonilla↗I-Chen Wu↗Ti-Rong Wu↗

arXiv:2606.17847v1 Announce Type: new Abstract: WallGo is a recently introduced strategic board game popularized by the 2025 Netflix series The Devil's Plan. Although played on a small 7 x 7 board, its combination of stone movement and wall placement yields high game-tree complexity and intricate strategic interactions. Despite its growing popularity, WallGo remains underexplored. This paper presents WallZero, an AlphaZero-based agent for the two-player WallGo setting. We introduce tailored action and feature designs to improve playing performance significantly. In the evaluation, WallZero defeats two professional Go players who participated in this study, securing on average 1.98x more territory per game. Beyond its strength, we use WallZero to assess game fairness and identify key strategies for mastering WallGo. Interestingly, our results show that the opening used in the Netflix series yields a more balanced game. Our code is available at https://rlg.iis.sinica.edu.tw/papers/wallzero.

Read & Discuss → View Source →

22.

arXiv (CS.CV) 2026-06-16 DOI: arXiv:2606.15763

The Circumplex Degeneracy Behind the Rare-Class Limit in Affect Recognition

Authors:

Van Thong Huynh↗Hong Hai Nguyen↗Soo-Hyung Kim↗

In-the-wild expression recognition persistently fails on a few rare emotions, and the standard explanation is class imbalance. Through a controlled multi-task study on two benchmarks, we show the failure is instead a property of affect geometry: the rare classes are degenerate on Russell's circumplex, and that degeneracy bounds what any loss or cost can achieve. Our instrument is a circumplex-cost optimal-transport term that prices expression confusions by their valence-arousal distance. The term improves the official score and expression macro-F1, but a control most studies omit shows the gain is not geometric: a uniform cost, equivalent to a generic confidence penalty, matches it on Aff-Wild2 (p=0.625) and significantly exceeds it on AffectNet (+0.057 over base, larger than the circumplex). What the geometry reshapes is the structure of the errors, making them affectively nearer the truth on Aff-Wild2 (p=0.031 against the uniform control), an effect that does not survive on AffectNet, where a visual confound at the far corner of the circumplex overwhelms it. The rare-class failure, by contrast, is stable across both datasets we examine: the degenerate pairs (anger-fear on Aff-Wild2, anger-contempt on AffectNet) resist frequency-based interventions, the transport term, and an action-unit-augmented cost built specifically to separate them. We conclude that progress on rare expressions requires representations that distinguish the classes, not supervision that reprices their confusions, and we provide the controls and metrics needed to tell the two apart.

Read & Discuss → View Source →

23.

arXiv (CS.CV) 2026-06-15 DOI: arXiv:2602.01801

Fast Autoregressive Video Diffusion and World Models with Temporal Cache Compression and Sparse Attention

Authors:

Dvir Samuel↗Issar Tzachor↗Matan Levy↗Michael Green↗Gal Chechik↗Rami Ben-Ari↗

Autoregressive video diffusion models enable streaming generation, opening the door to long-form synthesis, video world models, and interactive neural game engines. However, their core attention layers become a major bottleneck at inference time: as generation progresses, the KV cache grows, causing both increasing latency and escalating GPU memory, which in turn restricts usable temporal context and harms long-range consistency. In this work, we study redundancy in autoregressive video diffusion and identify three persistent sources: near-duplicate cached keys across frames, slowly evolving (largely semantic) queries/keys that make many attention computations redundant, and cross-attention over long prompts where only a small subset of tokens matters per frame. Building on these observations, we propose a unified, training-free attention framework (FAST-AR) for FAST-AutoRegressive diffusion, consisting of three components: TempCache compresses the KV cache via temporal correspondence to bound cache growth; AnnCA accelerates cross-attention by selecting frame-relevant prompt tokens using fast approximate nearest neighbor (ANN) matching; and AnnSA sparsifies self-attention by restricting each query to semantically matched keys, also using a lightweight ANN. Together, these modules reduce attention, compute, and memory and are compatible with existing autoregressive diffusion backbones and world models. Experiments demonstrate up to x5 - x10 end-to-end speedups while preserving near-identical visual quality and, crucially, maintaining stable throughput and nearly constant peak GPU memory usage over long rollouts, where prior methods progressively slow down and suffer from increasing memory usage.

Read & Discuss → View Source →

24.

arXiv (CS.CL) 2026-06-11 DOI: arXiv:2606.12234

On The Effectiveness-Fluency Trade-Off In LLM Conditioning: A Systematic Study

Authors:

Iuri Macocco↗Pau Rodr\'iguez↗Arno Blaas↗Luca Zappella↗Marco Baroni↗Xavier Suau↗

Controlling the output of Large Language Models (LLMs) is a central challenge for their reliable deployment, yet a clear understanding of the involved trade-offs remains elusive. Current approaches to conditioning are often evaluated with a narrow focus on their effectiveness at injecting or removing a target concept, neglecting generation quality. We systematically investigate a range of conditioning methods in both injection and removal scenarios. We find that efficient steering methods frequently achieve conditioning at a steep cost to fluency. Furthermore, we identify a critical yet previously overlooked interaction with the training paradigm: activation steering methods are far less effective on instruction-tuned models than on their base counterparts. Simple prompting and full-fledged supervised fine-tuning, on the other hand, are viable options for concept injection, but are not as good at concept removal. Finally, cheaply computed textual metrics highly correlate to costly LLM-as-judge scores, and provide insights on the behavior of conditioning methods.

Read & Discuss → View Source →

25.

arXiv (quant-ph) 2026-06-16 DOI: arXiv:2606.16418

A short proof of the modified Kretschmann-Schlingemann-Werner conjecture

Authors:

Satvik Singh↗

arXiv:2606.16418v1 Announce Type: new Abstract: Let $\Phi_1, \Phi_2 : \mathbb{M}_d(\mathbb{C})\to \mathbb{M}_n(\mathbb{C})$ be two quantum channels with respective Stinespring isometries $V_1, V_2 : \mathbb{C}^{d}\to \mathbb{C}^{n} \otimes \mathbb{C}^{m}$ on any common dilation space $\mathbb{C}^{m}$. We prove that there exists a unitary $U$ on $\mathbb{C}^{m}$ such that $\|V_1-({\bf1}\otimes U)V_2\|_\infty\leq\sqrt{2\|\Phi_1-\Phi_2\|_\diamond},$ thus resolving vom Ende's modification of the Kretschmann-Schlingemann-Werner conjecture in the affirmative.

Read & Discuss → View Source →