×

Academic Intelligence · Curated Daily

Explore the Frontier of Global Academia

AcademicHub aggregates real-time literature from top journals and preprint platforms. Build your personal research radar and let large language models compile cross-disciplinary analysis briefings automatically.

Authors: Lin Qu ×
Shuffle
01.
arXiv (quant-ph) 2026-06-19

Extracting the physical content of Liouvillian eigenmodes: Semiclassical quantization

arXiv:2606.20271v1 Announce Type: new Abstract: Unlike in closed quantum systems where individual energy eigenstates are understood as physical excitations, open quantum systems have distinct right and left eigenstates of the Liouvillian that decay with time and are difficult to interpret. Here we introduce a physically motivated quasiprobability measure combining the two types of eigenstates that interprets a Liouville eigenmode as a set of coherences. This coherence measure is intimately connected to the return probability and allows one to visualize the modes as quasiprobability distributions in a "doubled" phase space. Using this measure we show that, remarkably, an oscillator retains its quantized "orbits" in phase space for a large class of linear and nonlinear damping, thus providing a formulation of semiclassical quantization for open systems. The orbits have measurable dynamical signatures and are broadened in the presence of a thermal bath, similar to energy levels. For quadratic systems, our results yield an extension of the concept of invariant tori, which play a central role in Hamiltonian systems.

02.
arXiv (CS.CL) 2026-06-15

CoRe: A Continuously Reward-Finetuned LLM Query Rewriter for Multi-Stage Context-Aware Relevance in Web-Scale Video Search

LLM-based query rewriters in production face a tension: the training reward must reflect how the rewrite is consumed by the production ranker, yet the training procedure must be cheap enough to support continuous redeployment as data drifts. We present CoRe (Context Relevance), such a system, redeployed weekly for over five months in a major short-video search engine. Our reward uses the deployed multimodal relevance model as its source and a multiplicative ratio form mirroring the production fusion algebra, closing the simulation-production gap that offline reward proxies leave open. A semi-online Mixed Preference Optimization loop makes this reward affordable at multi-million-instance weekly scale: a DPO-style pairwise objective restricts the gradient pass to a small top-k/bottom-k subset of sampled trajectories, and a phase structure reduces trainer/inference-server parameter syncs from per-step to per-phase. An automated promotion gate over reward-like and stability metrics detected and recovered from a real reward-hacking incident in production. Rewriter output is consumed as parallel relevance signals at recall, rawrank, and finerank without displacing the original signals, bounding rewriter-failure blast radius. Online A/B from two sequential production launches, first deploying the rewriter at finerank, then extending consumption to recall and rawrank, delivers statistically significant reductions in change-query rate on rewrite-impacted queries, with all headline relevance and engagement metrics moving in the expected direction.

03.
arXiv (CS.CL) 2026-06-16

MosaicQuant: Inlier-Outlier Disaggregation for Unified 4-Bit LLM Quantization

4-bit quantization significantly reduces the memory footprint and accelerates the inference of large language models (LLMs). However, its limited bit-width representation struggles to faithfully capture both dense common values (inliers) and rare large-magnitude values (outliers), causing substantial accuracy degradation. Existing mixed-precision methods mitigate this by retaining outliers in high precision, but at the cost of breaking the uniformity of low-bit execution, introducing precision conversion and extra data movement that undermine practical speedup. We propose MosaicQuant, a unified 4-bit LLM quantization paradigm built on a novel principle of inlier–outlier disaggregation. Rather than elevating outlier precision, MosaicQuant quantizes the full weight matrix into a dense 4-bit base component, where inliers are captured faithfully while outlier are inevitably quantized. A sparse 4-bit residual component is then introduced to compensate for these quantization errors, selectively targeting the most error-critical weight blocks where output distortion is shown to be concentrated. However, a unified representation alone is insufficient, as naïvely executing the sparse residual as a separate kernel still breaks the unified low-bit inference pipeline. To bridge this gap, we introduce ZipperEngine, which fuses sparse block computation into the dense 4-bit GEMM kernel via an overlapped pipeline, unifying not only the representation but also the execution into a single coherent low-bit inference pipeline. Extensive experiments on LLaMA3 and Qwen3 demonstrate that MosaicQuant preserves near-FP16 accuracy while achieving up to $1.24\times$ speedup over the W16A16 baseline.

04.
arXiv (CS.AI) 2026-06-15

HierSVA: A Data Synthesis Pipeline, Dataset, and Benchmark for LLM-Driven Hierarchical Hardware Formal Verification

arXiv:2606.13706v1 Announce Type: cross Abstract: We present HierSVA, an integrated suite that combines a pipeline, dataset, and benchmark for LLM-driven hierarchical hardware formal verification. HierSVA-SP pairs an RTL preprocessing toolchain with an LLM-in-the-loop formal verification flow to produce reference SystemVerilog Assertions (SVA) on hierarchical RTL. Applying it to BaseJump STL yields HierSVA-DS, a dataset of 342 modules, with hierarchy metadata and depths 0–9, accompanied by a deep subset of 28 module-bug pairs with natural-language specifications and bug variants. HierSVA-B decomposes assertion quality into six metric axes: syntax correctness, assertion proof success rate, vacuity, specification faithfulness, mutation coverage, and formal core coverage. Applying HierSVA-B to twelve recent LLMs reveals three findings. First, the module-level compile rate is 67.1\%; among generated assertions in evaluable runs, 82.1\% prove non-vacuously, but the corresponding assertion sets detect only 70.2\% of eligible injected faults and cover 36.2\% of the formal core. Second, on 211 evaluable model–module entries in the deep subset, assertion sets flag buggy RTL with 0.87 recall, but 40\% of predicted-buggy outcomes are false positives on correct RTL, limiting precision to 0.60. Third, agentic mode improves S1-style provability and strength metrics, but gains plateau and oscillate. Codes and artifacts are available at \href{https://github.com/HierSVAAnon/HierSVACodeAndArtifacts}{https://github.com/HierSVAAnon/HierSVACodeAndArtifacts}. Dataset is available at \href{https://huggingface.co/datasets/AnonymousHierSVA/HierSVA}{https://huggingface.co/datasets/AnonymousHierSVA/HierSVA}.

05.
arXiv (CS.LG) 2026-06-16

Semantic DLM+: Improving Diffusion Language Models through Bias-variance Trade-off in Transition Kernel Design

arXiv:2606.15327v1 Announce Type: new Abstract: Diffusion Language Models (DLMs) have demonstrated strong scaling capacity as alternatives to autoregressive language models. However, their performance is highly sensitive to the choice of transition kernels, and poorly designed kernels can lead to issues like training instability, slow convergence, and biased sampling. In this paper, we study this sensitivity through a principled analysis of generalization error and identify three critical factors: asymptotic bias (difficulty in approximating the posterior distribution), exposure bias (error propagation during sampling), and optimization variance induced by kernel dispersion. We further compare different transition kernels: masking diffusion yields sparse and easier posterior-approximation targets, while uniform diffusion provides stronger sampling-side repair but induces harder approximation. Motivated by this trade-off, we revisit a previously overlooked variant, semantic DLM (SemDLM), where the transition kernel corrupts tokens to neighborhoods that are semantically similar. Our theory suggests that SemDLM can serve as a plausible middle ground by reducing the posterior approximation difficulty of uniform diffusion while retaining repair ability. However, we find that SemDLM suffers from a semantic basin problem, where sampling repeatedly stays within a semantic region and produces low-diversity text. To address this, we propose SemDLM+, which adds a global transition and a semantic-frequency penalty during sampling. Experiments on LM1B and OpenWebText show that SemDLM+ improves training dynamics and achieves competitive language modeling and generation quality with satisfactory diversity.

06.
arXiv (CS.CL) 2026-06-15

DLawBench: Evaluating LLMs Through Multi-Turn Legal Consultation

Lawyer-client consultation is a critical starting point for legal services. Effective legal assistance hinges on eliciting sufficient and truthful information from clients in order to devise strategies that best protect their interests. This task requires Large Language Models (LLMs) not only to perform robust legal reasoning, but also to strategically elicit material facts through multi-turn interactions and effectively guide clients with diverse personalities. Yet existing legal benchmarks overlook this interactive capability. To fill this gap, we introduce DLawBench, a diagnostic benchmark for real-world legal consultation. Drawing on realistic client behavior, we characterize lawyer-client interactions into four types: Cooperative, Dependent, Withdrawn, and Adversarial. Using dialogues grounded in real cases, DLawBench evaluates whether LLMs can effectively conduct legal consultation under realistic conditions. DLawBench comprises 461 cases from Chinese and U.S. law, 5,532 paired fact entries, 3,411 inquiry rubrics, and 3,348 issue-resolution rubrics, and evaluates 26 representative LLMs. Systematic experiments show substantial headroom: the best-performing model, GPT-5.5, achieves only 0.562 on consultation-grounded legal reasoning. More importantly, DLawBench exposes both sycophancy in legal consultation and a paradox: models perform worse when clients need guidance most.

07.
arXiv (CS.CL) 2026-06-11

Decoding Multimodal Cues: Unveiling the Implicit Meaning Behind Hateful Videos

Hateful videos have become prevalent on online platforms, highlighting an urgent need for effective detection. However, existing studies primarily focus on binary classification and fail to provide contextual rationales that reveal the implicit meanings behind these judgments, significantly undermining model explainability. To fill this gap, we aim to achieve explainable hateful video detection, enabling models to provide contextual rationales that integrate relevant evidence and logical reasoning alongside decisions. This approach can comprehensively enhance the understanding of video content and the explainability of the decision-making process. We first introduce two datasets, Ex-HateMM and Ex-ImpliHateVid, for explainable hateful video detection. Each dataset provides fine-grained annotations of multimodal harmful elements, along with contextual rationales. We then propose an Information Augmentation and Reasoning Enhancement (IARE) framework designed for explainable detection. The framework employs an information augmentation phase that leverages the multimodal chain-of-thought to integrate harmful elements, thereby enriching rationale evidence. Additionally, IARE incorporates a reasoning enhancement phase, in which Direct Preference Optimization guides the model toward correct reasoning paths and away from incorrect ones, thereby improving the logical coherence of its justifications. We conduct extensive experiments on the two datasets, comparing multiple baselines with our proposed IARE framework. The results demonstrate that IARE achieves state-of-the-art performance while also generating accurate rationales.

08.
arXiv (CS.CL) 2026-06-16

CoRA: Confidence-Rationale Alignment for Reliable Chain-of-Thought Reasoning

Chain-of-thought (CoT) reasoning can improve LLM performance, but high answer confidence may be misleading when the accompanying CoT rationale is plausible yet incomplete or poorly supported. We study confidence–rationale alignment: whether a model's confidence in its committed answer is justified by its generated rationale. We introduce a GRPO-based reinforcement learning framework that jointly rewards answer correctness, committed-answer probability, and rubric-based rationale support, where the rubric assesses grounding, coherence, task match, and connection to the selected answer without revealing the gold answer to the judge. Across MedQA, MathQA, and OpenBookQA using three open-weight LLMs, our method reduces the confidence–rationale alignment error by up to 26.51% compared with untuned checkpoints, SFT, and correctness-only GRPO, while maintaining competitive accuracy and often improving calibration. These results show that reliable CoT reasoning requires not only confident answers, but rationales that substantively support them.

09.
arXiv (quant-ph) 2026-06-19

Space-time duality approach to (inhomogeneous) integrable quenches

arXiv:2606.20445v1 Announce Type: cross Abstract: Characterising the universal aspects of non-equilibrium quantum many-body dynamics is one of the key goals of this century's physics research. Progress, however, is hindered by the lack of general theoretical frameworks for studying interacting quantum matter far from equilibrium. A recent breakthrough has been the realization that several key non-equilibrium quantities, such as the rate of growth of entanglement or the fluctuations of conserved charges within finite subsystems, can be related to equilibrium properties through a space-time duality that effectively exchanges the roles of space and time. This observation effectively enables the study of non-equilibrium phenomena using tools and concepts borrowed from equilibrium statistical mechanics and thermodynamics. A first proof of principle of this framework, dubbed space-time duality approach (SDA), was provided by interacting integrable systems, where thermodynamic properties can often be characterized exactly, while dynamical quantities typically remain beyond analytical reach. Subsequent developments, however, revealed that the SDA suffered from an intrinsic ambiguity, restricting its applicability to homogeneous quenches and to charge fluctuations arising from symmetric initial states. Here we resolve this ambiguity from first principles and derive closed-form predictions for entanglement growth and charge fluctuations after general quantum quenches. We benchmark our results against the exact analytical solution of the Rule 54 quantum cellular automaton and extensive TEBD simulations of the XXZ chain. Moreover we show that, when specialised to the entanglement entropy, our framework naturally reproduces the predictions of the quasiparticle picture.

10.
arXiv (CS.CL) 2026-06-12

SENTINEL: Failure-Driven Reinforcement Learning for Training Tool-Using Language Model Agents

Language model agents are increasingly effective in solving realistic tasks through multi-turn tool use. However, training reliable tool-using agents remains challenging in practice. While reinforcement learning provides an on-policy paradigm for improving agents from their own environment interactions, its effectiveness depends heavily on the training task distribution. When tasks are fixed before training, the task distribution can become increasingly mismatched with the policy's evolving capabilities, causing many rollouts to be spent on uninformative tasks. We propose SENTINEL, a failure-driven reinforcement learning framework that turns the Solver's rollout failures into targeted training tasks. SENTINEL follows a Controller–Proposer–Solver loop: the Controller analyzes failed trajectories and summarizes recurring error patterns, the Proposer generates executable tasks that stress these weaknesses, and the Solver is trained on the targeted tasks. On Tau2-Bench Retail with Qwen3-4B-Thinking-2507, SENTINEL improves Pass\^{}1 from 66.4 to 74.9 and outperforms RL on general synthetic tasks across Pass\^{}k metrics. These results demonstrate that model failures provide an effective and scalable source of targeted training signal for improving tool-using language model agents.

11.
arXiv (CS.LG) 2026-06-16

Inference-Time Decision Calibration for Temporal Classification

arXiv:2606.16034v1 Announce Type: new Abstract: Temporal classification errors are often treated as representation failures, but they can also arise from how available evidence is converted into decisions. This paper proposes a representation–calibration decomposition for temporal classification. We keep a trained native classifier frozen and separate two inference-time interventions: a conservative residual multi-scale branch that adds auxiliary logits to the native prediction, and a post-hoc branch-aware calibrator that recombines native and residual evidence at decision time. This design distinguishes missing temporal evidence from underused decision-level evidence without retraining the backbone. Across FI-2010, PTB-XL, UCI-HAR, MHEALTH, and HARTH, we find that gains are strongly regime-dependent. Residual multi-scale evidence is most useful in noisy or representation-limited settings, especially short-horizon FI-2010 and weaker recurrent backbones, while branch-aware calibration helps when native and auxiliary logits contain complementary evidence not fully exploited by the raw decision rule. Near-saturated settings show limited gains from either intervention. These results suggest that temporal classification should be understood not only as representation learning, but also as the problem of trusting, combining, and calibrating evidence from multiple views.

12.
arXiv (CS.CV) 2026-06-19

iSAGE: A Human-in-the-Loop Framework for Remote Sensing Semantic Segmentation via Sparse Point Supervision

Semantic segmentation in remote sensing requires costly pixel-level annotations, and nearly every problem demands a new dataset since models rarely transfer across sensors, platforms, or geographies. Existing human-in-the-loop frameworks expand sparse clicks into dense supervision via auxiliary machinery (pseudo-labels, propagation, CRFs, foundation-model prompts, auxiliary heads), all operating on the model's predictive distribution. A confidently wrong pixel is indistinguishable from a confidently correct one in that distribution by construction, so no rule reading it can separate the two; the distinguishing signal is external to the model. This paper hypothesizes that expert clicks targeting confident model errors, not arbitrary pixels, suffice to match dense supervision, with no expansion machinery. iSAGE (Iterative Sparse Annotation Guided by Expert) realizes this hypothesis on an integrated open-source platform, where an error-weighted loss amplifies the gradient at each click and the annotation record itself is the dataset, extensible, correctable, and auditable. Experiments use a minimum-effort regime: at most one labeled pixel per class per frame. On BsB Aerial, iSAGE recovers 97.2% of dense supervision (74.79% mIoU on 0.040% of pixels) with contrasting class dynamics: amorphous classes (permeable areas) saturate from the seed, while small classes (cars) require late-iteration effort. On ISPRS Vaihingen (external benchmark), iSAGE reaches 76.78% mIoU with 0.011% of pixels, matching the dense baseline (76.65%) and exceeding all published methods. Under the same pipeline, four output-reading mechanisms (oracle entropy across budgets 1–100x, pseudo-labels across thresholds 0.90–0.99, CRF-based propagation, uniform random) plateau 7.4 to 14.5 pp below iSAGE. Across 31 surveyed methods, iSAGE is the only iterative human-in-the-loop framework operating without auxiliary machinery.

13.
arXiv (CS.AI) 2026-06-16

Autonomous End-to-End SOH Prediction Services for Battery Systems via Temporal-Contrastive Representation Learning

arXiv:2606.16434v1 Announce Type: cross Abstract: Accurate state of health (SOH) estimation is a critical diagnostic service for lithium-ion battery management. However, reliance on labor-intensive manual feature engineering and opaque black-box models hinders scalable industrial deployment. To address this, we introduce TC-SOH: a modular, plug-and-play service architecture for autonomous, end-to-end SOH prediction. TC-SOH employs a temporal-contrastive mechanism and a cross-window prediction pretext task to extract degradation-relevant representations directly from raw operational data. To improve transparency, we connect model efficacy with representation diagnostics: visualization, sensitivity analysis, redundancy analysis, bidirectional probing, future-SOH probing, and temporal shuffling show that learned features overlap with selected expert descriptors while retaining additional SOH-relevant variation, and that ordered temporal context improves subsequent-SOH prediction. Across four public datasets, TC-SOH outperforms the considered physics-informed and data-driven baselines, reducing MAPE by 1.91 times and RMSE by 2.13 times.

14.
arXiv (CS.AI) 2026-06-11

Grounding Computer Use Agents on Human Demonstrations

arXiv:2511.07332v2 Announce Type: replace-cross Abstract: Building reliable computer-use agents requires grounding: accurately connecting natural language instructions to the correct on-screen elements. While large datasets exist for web and mobile interactions, high-quality resources for desktop environments are limited. To address this gap, we introduce GroundCUA, a large-scale desktop grounding dataset built from expert human demonstrations. It covers 87 applications across 12 categories and includes 56K screenshots, with every on-screen element carefully annotated for a total of over 3.56M human-verified annotations. From these demonstrations, we generate diverse instructions that capture a wide range of real-world tasks, providing high-quality data for model training. Using GroundCUA, we develop the GroundNext family of models that map instructions to their target UI elements. At both 3B and 7B scales, GroundNext achieves state-of-the-art results across five benchmarks using supervised fine-tuning, while requiring less than one-tenth the training data of prior work. Reinforcement learning post-training further improves performance, and when evaluated in an agentic setting on the OSWorld benchmark using o3 as planner, GroundNext attains comparable or superior results to models trained with substantially more data,. These results demonstrate the critical role of high-quality, expert-driven datasets in advancing general-purpose computer-use agents.

15.
arXiv (quant-ph) 2026-06-15

Implementation of two-qubit Rydberg operations on neutral Rb-87 atoms in systems with different intermediate states

arXiv:2606.13975v1 Announce Type: new Abstract: This work presents an experimental setup for implementing two-qubit operations on neutral atoms ($^{87}$Rb) with the possibility of using two different Rydberg excitation schemes. One of them uses 5P$_{1/2}$ as the intermediate level and applies the second-stage beam locally to the addressed atoms. The second scheme uses the 6P$_{3/2}$ level; in this scheme, the particles to be entangled are moved to a separate zone through which both Rydberg beams pass. The advantages and limitations of both schemes are analyzed. Based on numerical modeling performed with a Julia package developed by the authors, it is demonstrated that the spatial configuration has a greater effect on quantum-operation fidelity than the choice of intermediate level. An experimental implementation of the scheme using the 6P$_{3/2}$ level is demonstrated, making it possible to achieve a two-qubit operation fidelity of 94%.

16.
arXiv (CS.CV) 2026-06-19

Shape of Thought: Progressive Object Assembly via Visual Chain-of-Thought

Multimodal models for text-to-image generation have achieved strong visual fidelity, yet they remain brittle under compositional structural constraints, notably generative numeracy, attribute binding, and part-level relations. To address these challenges, we propose Shape-of-Thought (SoT), a visual CoT framework for process-supervised progressive shape assembly in the rendered 2D domain, without external engines at inference time. SoT trains a unified multimodal autoregressive model to generate interleaved textual plans and rendered intermediate states, helping the model capture shape-assembly logic without producing explicit geometric representations. Unlike text-only CoT, each decision is grounded in a rendered state, making counts, attachments, topology, and intermediate part-addition errors inspectable across the trajectory. To support this paradigm, we introduce SoT-26K, a large-scale dataset of grounded assembly traces derived from part-based CAD hierarchies, and T2S-CompBench, a benchmark for evaluating structural integrity and trace faithfulness. Fine-tuning on SoT-26K achieves 88.4% on component numeracy and 84.8% on structural topology, outperforming direct generation by +24.2 points on component numeracy and +19.3 points on structural topology. SoT establishes a transparent testbed for rendered-domain structure-aware generation. The code is available at https://github.com/yuhuo03/Shape-of-Thought.

17.
arXiv (CS.AI) 2026-06-17

From Brewing to Resolution: Tracing the Internal Lifecycle of Code Reasoning in LLMs

arXiv:2606.17648v1 Announce Type: new Abstract: Standard accuracy metrics cannot explain why LLMs handle variable tracking but fail on semantically equivalent loops. We study an internal lifecycle of code reasoning in which models first brew the answer, making it linearly recoverable many layers before it becomes self-decodable, and then diverge into one of four resolution outcomes: Resolved, Overprocessed, Misresolved, or Unresolved. Understanding this lifecycle matters because similar task accuracies can mask fundamentally different failure modes that surface-level evaluation cannot detect. We introduce a dual diagnostic framework pairing layer-wise linear probing with Context-Stripped Decoding (CSD) and apply it to six code-reasoning task families across 16 models spanning Qwen, Llama, and DeepSeek architectures. All four outcomes carry substantial mass in every task family: overall Resolved is only 41.5%, with multiple tasks below 30%. Controlled sweeps over structure, depth, and operators expose task-specific failure bottlenecks: Function Call Resolved plunges from 61.1% to 2.5% as call depth increases from one to three. Across architectures and scales, the brewing scaffold remains stable, with normalized brewing duration 24-42% across all 16 models, while resolution success varies with capability. This indicates that the scaffold is a stable empirical regularity across the tested decoder-only Transformer families, whereas resolution success covaries with capability, scale, and training. Code: https://github.com/euyis1019/llm-brewing

18.
arXiv (quant-ph) 2026-06-16

Quantum Nonlocal Games on Graph Ensembles

arXiv:2606.16784v1 Announce Type: new Abstract: Quantum entanglement is one of the most striking discoveries in all of science. This effect allows, for instance, two spatially separated agents to coordinate their actions, without communication, to an extent that is both counter-intuitive, and provably impossible by any other physical means. A recently discovered example is that of mobile agents (players) performing spatial coordination tasks such as rendezvous, where the agents aim to meet on a network without communication. Until now, demonstrations of this advantage have relied on highly idealized conditions: agents are assumed to have complete knowledge of the topography, and experiments have been restricted to simulations using data generated by qubits within a single quantum processor. Here we address both limitations by developing a theory for graph ensembles that capture topographical uncertainty and by experimentally demonstrating the advantage in rendezvous scenarios between physically separated ion-trap systems with access to remote entanglement. Moreover, we simulate a broader set of problems on superconducting hardware. Surprisingly, when players are given the ability to gather more local information the quantum advantage increases – a feat impossible by classical means. Our findings establish a concrete route toward practical quantum advantages in motion coordination problems. More broadly, they point to a new way of using portable quantum devices to enhance collective decision-making in uncertain environments.

19.
arXiv (CS.CV) 2026-06-17

BusterX++: Towards Unified Cross-Modal AI-Generated Content Detection and Explanation with MLLM

The rapid advancement of generative AI has substantially improved image and video synthesis, amplifying the risk of multimodal visual misinformation. Recent MLLMs have shown promise for transparent AI-generated content detection through reasoning and explanation, yet existing approaches largely treat image and video forensics as isolated tasks, leaving cross-modal synergies underexplored. To address this, we present BusterX++, a unified MLLM for joint image and video detection with interpretable reasoning. We also introduce GenBuster-Bench++, a meticulously curated, difficulty-aligned benchmark containing balanced image and video samples spanning recent generation models and diverse real-world scenarios. Using this controlled setting, we revisit the widely adopted $SFT \rightarrow RL$ post-training paradigm. Notably, our findings demonstrate that a single-stage, pure RL strategy driven strictly by sparse outcome rewards consistently matches or surpasses a strong SFT+RL baseline across both unified and single-modality settings. Our key insight reveals that SFT imposes lower policy entropy, which restricts the policy search space and dampens exploratory freedom. In contrast, single-stage pure RL maintains higher policy entropy throughout training, effectively unlocking the spontaneous emergence of cross-modal capability transfer between image and video forensics. Extensive experiments demonstrate that BusterX++ achieves state-of-the-art performance, highlighting the powerful potential of RL for unified cross-modal visual reasoning.

20.
arXiv (CS.AI) 2026-06-16

RollArt: Disaggregated Multi-Task Agentic RL Training at Scale

arXiv:2512.22560v2 Announce Type: replace-cross Abstract: Agentic Reinforcement Learning (RL) trains LLMs through multi-turn interactions with environments, producing workloads that mix compute-bound prefill, bandwidth-bound decoding, CPU-heavy environment execution, and bursty reward evaluation. Existing systems either colocate all stages on a single GPU cluster or decouple them only at a coarse granularity, overlooking hardware heterogeneity and incurring substantial synchronization overhead across stages. We present ROLLART, a system for multi-task agentic RL on disaggregated infrastructure. ROLLART maps each pipeline stage to best-fit hardware, routing prefill-heavy tasks to compute-optimized GPUs, decode-heavy tasks to bandwidth-optimized GPUs, and environments to CPU clusters. It decouples rollout at the trajectory level, allowing generation, environment interaction, and reward scoring to proceed independently, so that slow or failed environments never block the others. ROLLART offloads stateless reward computation to serverless infrastructure and overlaps rollout with training via staleness-bounded asynchronous weight synchronization. Our results demonstrate that ROLLART effectively improves training throughput and achieves 1.31–2.05 \(\times\) training time reduction compared to various RL systems. We also evaluated ROLLART by training a hundreds-of-billions-parameter MoE model for Qoder product on an Alibaba cluster with above 3,000 GPUs, demonstrating its stability and scalability.

21.
arXiv (CS.CL) 2026-06-19

Vero: An Open RL Recipe for General Visual Reasoning

What does it take to build a visual reasoner that works across charts, science, spatial understanding, and open-ended tasks? The strongest vision-language models (VLMs) suggest that broad visual reasoning is within reach, yet their closed data and reinforcement learning (RL) pipelines make their gains difficult to study, reproduce, or extend. We introduce Vero, a family of fully open VLMs that match or exceed existing open-weight models across diverse visual reasoning tasks. We scale RL data and rewards across six broad task categories, constructing Vero-600K, a 600K-sample dataset from 59 datasets, and designing task-routed rewards that handle heterogeneous answers. Across VeroEval, our 30-benchmark suite, Vero-600K outperforms existing RL datasets under controlled comparisons. Applied to five starting models, Vero variants gain 2.9-5.4 points on average over their initial models. Notably, Vero-Qwen3I-8B, trained on the Instruct model, surpasses Qwen3-VL-8B-Thinking by 3.8 points on average without additional distillation. Systematic ablations reveal that different task categories elicit distinct reasoning patterns and that broad gains depend on learning them jointly rather than in isolation. All data, code, and models are publicly available.

22.
arXiv (CS.CL) 2026-06-19

Where Does Social Reasoning Come From? Capability Provenance in Language Models

We use training-data attribution as an interpretable tool for capability discovery, mapping which regions of the pretraining corpus support social-reasoning versus STEM-reasoning in OLMo3-7B. Training-data attribution measures how strongly each training document influences a model's predictions on a benchmark, but document-level scores are too noisy to identify which corpus regions support which capabilities, and prior work has emphasized factual knowledge rather than reasoning. We compute gradient-based attribution (TrackStar via Bergson) over a working set drawn from the de-duplicated Dolma3 mix, aggregate influence across WebOrganizer's 24-format x 24-topic taxonomy (576 bins), and contrast benchmark pairs in a 2x2 design that varies domain (social vs. STEM) and capability type (reasoning vs. knowledge): SocialIQA and MMLU Social Sciences against ARC-Challenge and MMLU STEM. Social and STEM reasoning draw on qualitatively distinct corpus regions, and the contrast is sharper at the reasoning level than at the knowledge level. Targeted machine unlearning provides partial causal validation: forgetting high-attribution topic bins (e.g., Literature for SocialIQA) degrades the aligned benchmark more than within-bin random baselines, and we open-source all code, sampling manifests, the bin-level influence matrix, and unlearning checkpoints.

23.
arXiv (CS.AI) 2026-06-19

Beyond Static Leaderboards: Predictive Validity for the Evaluation of LLM Agents

arXiv:2606.19704v1 Announce Type: new Abstract: Agent benchmarks are growing fast, but no single benchmark touches more than four or five of the dimensions that deployment exposes. This paper aggregates the largest coordinated deep-dive of one MCP-based industrial-agent benchmark to date: fourteen parallel implementation studies covering new asset classes (including a multi-modal visual extension), alternative orchestrations, retrieval strategies, reasoning modes, infrastructure optimizations, and evaluation-methodology probes. Consolidating those studies with seven prior agent benchmarks, we argue that aggregate-score leaderboards systematically underspecify deployed-agent evaluation. Rankings derived from aggregate scores do not transfer to out-of-distribution settings; recent public-to-hidden competition retrospectives provide direct empirical evidence of this rank instability. We propose ranking configurations by predictive validity, the correlation between in-sample and out-of-sample rank, rather than in-sample mean, and report a twelve-tier measurement apparatus that exposes the deployment-relevant dimensions HELM and its agent-era successors collapse. The position is operationalized through three falsifiable out-of-distribution criteria with explicit thresholds; existing evidence partly supports it but is too thin to confirm. We close with a pre-registered pilot design and a field-level vision for what the next generation of agentic benchmarks should report.

24.
arXiv (CS.CV) 2026-06-16

Structure-aware Knowledge-guided Heterogeneous Mamba for Zygomaticomaxillary Suture Assessment

The Zygomaticomaxillary Suture is a key circummaxillary structure that connects the zygomatic bone and the maxilla, which serves as a primary site of resistance during maxillary advancement, and its maturation status directly influences the timing and efficacy of orthopedic interventions. However, accurate staging of ZMS maturation remains challenging due to subtle high-frequency transitions in suture lines and the global semantic ambiguity between adjacent stages. To address this, we present the first public ZMS dataset, comprising 3,790 ZMS images covering the entire age range from 4 to 24 years. Based on this dataset, we propose SKMamba, a Structure-aware and Knowledge-guided Mamba-based multi-modal framework for automated ZMS maturation assessment. SKMamba adopts a decoupled dual-path architecture that mimics the hierarchical diagnostic process used by experienced orthodontists. We first introduce an Implicit Edge Extractor (IEE), which leverages structural pre-training to reduce trabecular noise and accentuate sutural boundaries. Complementarily, a Cross-Modal Semantic Alignment (CSA) module is designed to incorporate anatomical descriptions from a large language model (LLM). This module helps align local morphological cues with global semantic descriptions while ensuring that objective morphological evidence remains the primary basis for decisions. Extensive experiments on our ZMS dataset demonstrate that SKMamba achieves state-of-the-art performance compared to existing methods. Code is available at https://github.com/galaxygxq1116/SKMamba.

25.
arXiv (CS.AI) 2026-06-16

LabOSBench: Benchmarking Computer Use Agents for Scientific Instrument Control

arXiv:2606.16802v1 Announce Type: new Abstract: Current computer-use benchmarks primarily focus on software operation tasks in virtualized systems, whereas scientific instrumentation scenarios require coordinated control over complex interfaces, and feedback-driven parameter adjustment. However, directly evaluating agents on physical high-precision instruments is impractical due to high cost, safety risks, limited accessibility, and difficulty in ensuring reproducible evaluation. This motivates the need for a simulated yet realistic testbed that preserves the operational challenges of scientific instruments while enabling scalable and safe benchmarking. To this end, we introduce LabOSBench, a challenging benchmark for multimodal GUI agents built on a suite of web-based scientific-instrument simulators. Operating directly via a browser, LabOSBench avoids resource-heavy OS virtualization while supporting flexible task configuration and execution-based evaluation. Specifically, LabOSBench constructs 96 subtasks across eight instrument simulators, covering workflows from sample loading, alignment, parameter tuning, and data acquisition to result inspection. We evaluate general-purpose vision-language models, specialized GUI agent models, and advanced agentic frameworks at both subtask and end-to-end levels. Our experiments reveal that while existing agents can complete many structured GUI subtasks, they still struggle with feedback-driven operations and long-horizon workflow execution. Overall, LabOSBench provides a reproducible, low-cost testbed for advancing computer-using agents toward scientific-instrument control.