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.

01.
arXiv (CS.CL) 2026-06-12

Reasoning Models Know What's Important, and Encode It in Their Activations

Language models often solve complex tasks by generating long reasoning chains, consisting of many steps with varying importance. While some steps are crucial for generating the final answer, others are removable. Determining which steps matter most, and why, remains an open question central to understanding how models process reasoning. We investigate if this question is best approached through model internals or through tokens of the reasoning chain itself. We find that model activations contain more information than tokens for identifying important reasoning steps. Crucially, by training probes on model activations to predict importance, we show that models encode an internal representation of step importance, even prior to the generation of subsequent steps. The internal representations of importance in different models yield high agreement on which steps are important. The representation is distributed across layers, and does not correlate with surface-level features, such as a step's relative position or its length. Our findings suggest that analyzing activations can reveal aspects of reasoning that surface-level approaches fundamentally miss, indicating that reasoning analyses should look into model internals.

02.
arXiv (CS.CL) 2026-06-12

Understanding helpfulness and harmless tension in reward models

Reward models are a key component of reinforcement learning from human feedback (RLHF), aligning language models toward both helpful and harmless behaviour. However, the internal mechanisms underlying these objectives and their conflicts remain poorly understood. We study alignment tension in reward models trained under helpfulness-only, harmlessness-only, and mixed-objective settings. We find that mixed-objective models often underperform single-objective models, indicating interference between objectives. Using activation-based methods, we identify neurons associated with each objective and study their functional roles via targeted ablations. We find that these neurons causally support their corresponding objectives while often negatively affecting the opposing one. We find that a substantial proportion of neurons are shared between helpfulness and harmlessness, and that these shared neurons exert a disproportionate influence on model behaviour, contributing to alignment tension. Additionally, our results provide insights and mechanistic interpretation into how alignment objectives are represented in reward models and why multi-objective alignment remains challenging, motivating future work on disentangled and controllable alignment methods.

03.
arXiv (CS.LG) 2026-06-19

3D-DLP: Self-Supervised 3D Object-Centric Scene Representation Learning

arXiv:2606.19451v1 Announce Type: new Abstract: We introduce 3D-DLP, a self-supervised object-centric representation learning model that decomposes scene-level RGB-D or voxel observations into a set of 3D latent particles. Building on the Deep Latent Particles (DLP) framework, each particle encodes disentangled attributes, including 3D keypoint position, bounding box dimensions, and appearance features, and represents a distinct entity in the scene. The model learns interpretable per-particle segmentation maps through an end-to-end self-supervised reconstruction objective. We demonstrate on both simulated and real-world datasets that the learned latent space is interpretable and controllable: by manipulating particle positions and decoding, we can generate novel scene configurations. Furthermore, we show that leveraging these compact 3D latent particles for downstream robotic manipulation improves performance over baselines that either lack explicit 3D information or rely on memory-intensive dense 3D inputs without object-centric structure. Code and videos are available at https://eubooks3003.github.io/3d-dlp.

04.
arXiv (CS.CL) 2026-06-12

ArogyaSutra: A Multi-Agent Framework for Multimodal Medical Reasoning in Indic Languages

Multimodal Large Language Models (MLLMs) have shown promising reasoning capabilities in general domains, yet their performance remains limited in specialized settings such as healthcare, especially in multilingual and low-resource scenarios. This gap is critical in regions like rural India, where patients often express complex medical queries in native Indic languages and rely on multimodal inputs such as medical images. Existing English-centric MLLMs struggle to support such use cases, limiting equitable access to AI-driven healthcare assistance. To address this challenge, we introduce ArogyaBodha, a large-scale multilingual multimodal medical question-answer dataset constructed from eight heterogeneous sources, covering 31 body systems, six imaging modalities, and 21 clinical domains across English and seven major Indian languages. We further propose ArogyaSutra, an actor-critic-based multi-agent framework that integrates tool grounding with dual-memory mechanisms for step-wise, reasoning-aware decision making, and uses stored actor-critic simulation trajectories for distillation. Experiments show that our dataset and framework improve multilingual medical reasoning accuracy across all Indic languages, with ablations validating the contribution of each component. The source code and dataset are available at: https://iitp-cse.github.io/ ArogyaSutra/

05.
medRxiv (Medicine) 2026-06-22

MinderCare: protocol for a mixed-methods evaluation of a digitally enabled dementia care service.

Introduction and aims Dementia is a growing public health challenge affecting millions of people worldwide. It is a progressive condition that increases the risk of infections, falls, hospital admissions, dependence in activities of daily living, safety issues such as wandering, care home transfers, and death. New ways of supporting people living with dementia (PLWD) at home are urgently needed. We describe the MinderCare study which evaluates a digitally enabled care model that integrates low-burden sensor-based remote monitoring within a nurse-led clinical service. Methods and analysis In this mixed-methods study, we will recruit 100 people with confirmed or suspected dementia living at home and deploy the Minder remote monitoring system for at least 12 months. A detailed characterisation of the cohort will be obtained, including cognition, frailty, participant and carer wellbeing, functioning, and quality of life. The feasibility, acceptability, sustainability, and resource requirements of the service will also be assessed. Low-cost sensors provide information about behaviour, environment and physiology from the home. Machine-learning algorithms have been used to develop digital biomarkers of infection, sleep, night-time behaviours, daily activities and routines, and the effects of clinical events and treatment. These will be assessed through clinical reports of sensor-derived data that include anomaly alerts provided to the clinical teams. Algorithms will be assessed for their clinical utility and acceptability. The comparative-effectiveness component will be designed as a target trial emulation using linked electronic health-record data to construct a time-indexed external usual-care control cohort. The primary comparative outcome will be Days Alive and Out of Hospital (DAOH) over 12 months from the activation-index date, with healthcare utilisation, costs, institutionalisation and mortality assessed as secondary outcomes. DAOH and estimated MinderCare effects will also be examined across prespecified strata of baseline inpatient utilisation. Ethics and dissemination Ethical approval has been granted by the North East Newcastle and North Tyneside 2 Research Ethics Committee, and the study has received confirmation of capacity and capability by the Imperial College Healthcare NHS Trust. Study findings will be disseminated to patients, health and social care professionals, and policymakers through peer-reviewed publications and conference presentations. Study registration number: ISRCTN14997677 and NIHR portfolio CPMSID 63023.

06.
bioRxiv (Bioinfo) 2026-06-21

OracleScreen-LILRB4: Machine Learning-Guided Discovery of Myeloid Immune Checkpoint Binders Validated in Patient-Derived Cells

The identification of small molecule modulators of immune checkpoint proteins remains a significant challenge in drug discovery due to the flat, featureless nature of protein-protein interaction interfaces and the characteristically low hit rates observed in conventional high-throughput screening campaigns. Here we report OracleScreen-LILRB4, an ensemble machine learning framework trained on quantitative biophysical screening data from two structurally diverse compound libraries (19,800 compounds total) screened against the myeloid immune checkpoint leukocyte immunoglobulin-like receptor B4 (LILRB4/ILT3). By formulating binding prediction as a regression task targeting continuous {Delta}Fnorm values rather than binary hit classifications, OracleScreen-LILRB4 achieved a mean Spearman R of 0.61 and ROC-AUC of 0.86 under scaffold-aware cross-validation. Prospective virtual screening of a 45,760-member compound library and experimental validation of the top 200 predictions yielded a 28.5% hit rate, representing a 15.0-fold enrichment over baseline, with 16 compounds demonstrating nanomolar-affinity LILRB4 (ILT3) engagement. Lead compounds ORS-22 and ORS-14 restored anti-tumor immune activity across patient-derived colorectal cancer and acute myeloid leukemia co-culture systems, reversing SCG2-mediated immunosuppression and recovering cytotoxic T-cell function. These findings establish OracleScreen-LILRB4 as an effective computational framework for accelerating small molecule discovery against non-enzymatic immune checkpoint targets.

07.
arXiv (CS.CV) 2026-06-19

GH-ESD: Grounded Hypothesis-Driven Error Slice Discovery for Instance-Level Vision Tasks

Systematic failures of vision models on semantically coherent subsets, known as error slices, reveal limitations in robustness and evaluation. Existing slice discovery approaches largely model slices as clusters in representation space or combinations of predefined attributes. While effective for image-level classification, such formulations are insufficient for instance-level tasks such as object detection and segmentation, where failures often arise from contextual relational and spatially grounded visual patterns. We propose GH-ESD (Grounded Hypothesis-Driven Error Slice Discovery), a generate and verify framework that reformulates slice discovery as grounded hypothesis generation and statistical verification. GH-ESD constructs relational failure hypotheses using LLM priors and grounded visual evidence, discovers hypothesis slices at the instance level via Vision Language Models, and verifies them through statistical trend analysis over instance-level errors. We also introduce GESD (Grounded Error Slice Dataset), a new benchmark for instance-level error slice discovery, providing expert-defined and spatially grounded slices derived from detection and segmentation failures. Extensive experiments demonstrate that GH-ESD consistently outperforms baselines, improving Precision@10 by 0.10 (0.73 vs. 0.63) on the GESD benchmark for detection tasks, while also supporting segmentation scenarios. GH-ESD identifies interpretable slices that facilitate actionable model improvements. The GESD dataset will be made publicly available upon acceptance.

08.
medRxiv (Medicine) 2026-06-18

MOSAIC: Methylation-Oriented Site Analysis and Information Classifier for Robust Epigenomic Classification of Acute Leukemia in Clinical Cohorts with Variable Tumor Purity

DNA methylation-based classification offers a rapid diagnostic complement to conventional molecular workflows in acute leukemia. Existing classifiers are trained on array-derived reference cohorts whose construction favors specimens with adequate tumor content, leaving clinically relevant low-purity specimens underrepresented and classifier robustness in this regime uncharacterized. On held-out low-purity specimens, existing classifiers were concordant with expert pathology in only 7 of 10 (MARLIN) and 5 of 10 (ALMA) cases, motivating a classifier built to maintain accuracy at low tumor purity. We developed MOSAIC (Methylation-Oriented Site Analysis and Information Classifier), a neural network classifier built to maintain accuracy across the full range of tumor purities encountered in clinical practice. MOSAIC is a neural network trained on publicly available array-based methylation data augmented with native methylation calls from Oxford Nanopore sequencing. MOSAIC was evaluated on low-purity specimens held out entirely from training. On these held-out low-blast leukemia specimens, all below 25% blasts and including a case at 1.4%, MOSAIC was concordant with expert pathology in every case, recovering the correct subtype where diluted disease signal would otherwise be mistaken for normal or unrelated tissue. Gradient-based saliency analysis showed that the network relies on a partially distinct set of discriminative CpG probes when classifying low-blast specimens. MOSAIC demonstrates that augmenting training with clinically representative clinical specimens yields methylation-based leukemia classification that maintains effectiveness under the variable tumor purity of real clinical cohorts.

09.
arXiv (CS.CL) 2026-06-16

Beyond Text-to-SQL: An Agentic LLM System for Governed Enterprise Analytics APIs

Enterprise analytics aims to make organizational data accessible for decision-making, yet non-technical users still face barriers when using traditional business intelligence tools or Text-to-SQL systems. While recent Text-to-SQL approaches based on Large Language Models (LLMs) promise natural language access to structured data, they fall short in enterprise settings where analytics pipelines rely on governed APIs rather than raw databases. In practice, these APIs encapsulate complex business logic to ensure consistency, auditability, and security. However, delegating mathematical or aggregation logic to an LLM introduces reliability and compliance risks. To this end, we present Analytic Agent, an LLM-based agentic system that translates natural language intents into secure interactions with enterprise analytics APIs. Evaluated on 90 real enterprise use cases constructed by domain experts, it reliably interprets user goals, validates permissions, executes governed queries, and generates compliant visualizations through multi-step reasoning and policy-aware orchestration.

10.
arXiv (CS.LG) 2026-06-19

Quantile of Means: A Bonus-Free Ensemble Method for Minimax Optimal Reinforcement Learning

arXiv:2606.20107v1 Announce Type: new Abstract: Optimal Reinforcement Learning (RL) algorithms typically rely on carefully constructed count-based uncertainty estimates to drive exploration. Although theoretically sound, such estimates are hard to compute in practical settings and therefore offer limited insight for designing exploration heuristics. Meanwhile, ensembling has emerged as a practical approach, but remains without theoretical justification. Building on a recent ensemble-based method for Multi-Armed Bandits, we propose a quantile-based ensemble method for finite-horizon Markov Decision Processes (MDPs). Our simple count-free approach achieves optimal variance-dependent regret bounds, providing theoretical grounding for ensemble-based exploration in RL.

11.
arXiv (CS.LG) 2026-06-18

CODEBLOCK: Learning to Supervise Code at the Right Granularity

arXiv:2606.18286v1 Announce Type: new Abstract: Supervised fine-tuning of code LLMs typically applies uniform cross-entropy loss to all response tokens, implicitly assuming that every token provides equally useful learning signal. Recent token-level selection methods challenge this assumption in natural-language SFT by supervising only high-value tokens. However, directly transferring token-level masking to code can break syntactically and semantically coherent program units, because code depends on structural completeness and definition-use relations. We therefore propose CodeBlock, a structure-aware sparse supervision framework that selects structure-complete code evidence rather than isolated tokens. CodeBlock first selects high-quality instruction-response pairs, then partitions code responses into syntactically coherent coding items, estimates their utility by aggregating generalized cross-entropy over core logic tokens, and reranks them with data-flow reach and bridge signals to prioritize blocks that propagate or connect important program dependencies. During training, the full response remains available as context, while loss is applied only to selected code items and informative natural-language tokens. Experiments on six code-generation benchmarks show that CodeBlock achieves stronger average pass@1 than full-token SFT and competitive selection baselines, while using only 1.9% of supervised response tokens.

12.
arXiv (CS.AI) 2026-06-11

MoCA-Agent: A Market-of-Claims Code Agent for Financial and Numerical Reasoning

arXiv:2606.11537v1 Announce Type: new Abstract: Financial and tabular question answering requires more than fluent reasoning: answers must be grounded in the exact facts, formulas, units, signs, and scales that support them. A single misread cell or incorrect operation can silently produce a plausible but wrong result. We introduce \textsc{MOCA-Agent}, a market-of-claims code agent that replaces free-form multi-agent debate with claim-level verification. The system decomposes each question into typed atomic claims, asks specialist trader agents to buy or sell those claims, clears their orders into confidence-weighted accept/reject decisions, and synthesizes an executable Python program from market-supported evidence. A code-aware verifier then checks the program for execution, structural consistency, and common financial reasoning errors, with at most one market-aware repair round. Across ten public benchmarks spanning financial numerical reasoning, general tabular reasoning, ESG question answering, and multimodal chart reasoning, \textsc{MOCA-Agent} achieves strong performance using a fixed Qwen3.6-27B backbone, including $78.3\%$ on FinQA, $76.0\%$ on FinanceMath, $71.2\%$ on MultiHiertt, $86.9\%$ on ESGenius, and $85.6\%$ average on FinChart-Bench. These results show that aggregating evidence at the level of atomic claims, rather than whole answers, improves robustness in high-stakes numerical reasoning.\footnote{The code and data are available: https://github.com/UBC-NLP/MoCA-Agent.

13.
arXiv (CS.AI) 2026-06-12

Free-Placement Optimization of Ground Station Locations for Low-Earth Orbit Satellites

arXiv:2606.12667v1 Announce Type: cross Abstract: Rapidly expanding low Earth orbit satellite constellations are placing increasing demands on terrestrial ground networks, motivating the development of more efficient ground station network designs. Current approaches select sites from predefined locations, limiting optimization to existing infrastructure and constraining performance. In contrast, free-placement optimization operates over a continuous spatial domain on Earth, broadening the search space and allowing higher-throughput configurations at the cost of potentially requiring new infrastructure deployment. In this work, we introduce SCORE (Sequential Cyclic Optimization via Refinement & Evaluation), a two-stage free-placement method for ground station design. SCORE combines sequential coordinate selection with cyclic refinement to manage high-dimensionality, non-convexity, and local minima that challenge global optimizers. We benchmark SCORE against one-shot methods such as differential evolution (DE) and integer programming approaches using locations from Kongsberg Satellite Services and the World Teleport Association. Tests across two commercial Earth observation constellations (Capella Space and ICEYE) and one synthetic Walker-Star constellation show that SCORE requires up to 5x fewer function evaluations to converge relative to DE while improving downlink throughput by up to 13%. Compared to fixed-site methods, unconstrained SCORE achieves up to 15% greater total downlink, establishing a strong empirical performance benchmark for flexible placement; infrastructure-constrained SCORE retains over 92% of this gain while restricting placement to within proximity of existing fiber and power infrastructure. We also explore trade-offs between expanding existing stations and deploying new sites, informing future ground network design for operational constellations.

14.
arXiv (quant-ph) 2026-06-17

Practical Tests and Witnesses of Fermionic non-Gaussianity

arXiv:2605.26218v2 Announce Type: replace Abstract: Fermionic Gaussian states describe free fermions and underlie the mean-field picture of matter, from metals to superconductors; they are also efficiently simulable on classical computers. Departures from Gaussianity – the correlations produced by interactions – are therefore what make a fermionic system hard to simulate classically and useful for quantum computation, analogous to the role of magic in stabilizer-based quantum computation. Yet detecting and quantifying such non-Gaussianity at scale has remained challenging. Here we introduce practical tests and witnesses of fermionic non-Gaussianity built on fermionic antiflatness, a measure derived from the two-point covariance matrix. We estimate it with two protocols – a two-copy Bell measurement and a single-copy scheme using commuting Majorana bilinears – that determine whether a state is Gaussian or far from it at lower measurement cost than existing approaches, using only operations native to fault-tolerant hardware. For mixed states, a purity-corrected witness certifies non-Gaussianity and remains robust under strong noise; running it on the IQM quantum processor, we find that noise can both reduce and enhance non-Gaussianity. Finally, we show that preparing pseudorandom fermionic states requires extensive non-Gaussianity. Together, these tools enable the study and certification of non-Gaussian fermionic resources on present-day quantum devices.

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

Denoising Score Matching with Random Features: Insights on Diffusion Models from Precise Learning Curves

arXiv:2502.00336v3 Announce Type: replace Abstract: We theoretically investigate the phenomena of generalization and memorization in diffusion models. Empirical studies suggest that these phenomena are influenced by model complexity and the size of the training dataset. In our experiments, we further observe that the number of noise samples per data sample ($m$) used during Denoising Score Matching (DSM) plays a significant and non-trivial role. We capture these behaviors and shed insights into their mechanisms by deriving asymptotically precise expressions for test and train errors of DSM under a simple theoretical setting. The score function is parameterized by random features neural networks, with the target distribution being $d$-dimensional Gaussian. We operate in a regime where the dimension $d$, number of data samples $n$, and number of features $p$ tend to infinity while keeping the ratios $\psi_n=\frac{n}{d}$ and $\psi_p=\frac{p}{d}$ fixed. By characterizing the test and train errors, we identify regimes of generalization and memorization as a function of $\psi_n,\psi_p$, and $m$. Our theoretical findings are consistent with the empirical observations.

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

Remember, Don't Re-read: Stateful ReAct Agents for Token-Efficient Autonomous Experimentation

arXiv:2606.14945v1 Announce Type: new Abstract: The autoresearch pattern enables autonomous experimentation by having a large language model (LLM) iteratively modify code to optimize a target metric. Its stateless design, however, reconstructs experimental context from scratch at every iteration, incurring $O(n)$ token cost per iteration and $O(n^{2})$ total. This work reformulates the pattern as a stateful ReAct agent using LangGraph, where typed persistent state carries experimental history across iterations via a tool-calling interface. Two benchmarks are evaluated: hyperparameter tuning (15 iterations, small per-iteration observations) and code performance optimization (40 iterations, large per-iteration observations containing full source code and benchmark results). On hyperparameter tuning, the stateful agent consumes 90\% fewer tokens (2{,}492 vs.\ 24{,}465). On code optimization, the stateful agent consumes 52\% fewer tokens (627K vs.\ 1{,}275K) while achieving comparable optimization quality on both tasks. The token reduction is structural: the stateless agent re-reads the full history at $O(n)$ cost per iteration, while the stateful agent operates within a fixed-size conversation window at $O(1)$ cost. This paper describes the architecture in sufficient detail for practitioners to implement a stateful autoresearch agent for their own workflows.

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

DVD: Discrete Voxel Diffusion for 3D Generation and Editing

We introduce Discrete Voxel Diffusion (DVD), a discrete diffusion framework to generate, assess, and edit sparse voxels for SLat (Structured LATent) based 3D generative pipelines. Although discrete diffusion has not generally displaced continuous diffusion in image-like generation, we show that it can be an effective first-stage prior for sparse voxel scaffolds. By treating voxel occupancy as a native discrete variable, DVD avoids continuous-to-discrete thresholding and provides a simple framework for voxel generation, uncertainty estimation, and editing. Beyond quality gains, DVD provides more interpretable generation dynamics through explicit categorical modeling. Furthermore, we leverage the predictive entropy as a robust uncertainty metric to identify ambiguous voxel regions and complicated samples, facilitating tasks such as data filtering and quality assessment. Finally, we propose a lightweight fine-tuning strategy using block-structured perturbation patterns. This approach empowers the model to inpaint and edit voxels within a single sampling round, requiring negligible auxiliary computation and no additional model evaluations. Code is available at https://github.com/TeCai/DVD.

18.
arXiv (CS.AI) 2026-06-18

Structured Cognitive Loop for Behavioral Intelligence in Large Language Model Agents (Extended Revision: From Behavioral Architecture to Epistemic Accountability)

Authors:

arXiv:2510.05107v5 Announce Type: replace Abstract: The central challenge for AI agents is not only performance but accountability. Agents that act through opaque prompt sequences may produce correct outputs, but they provide little basis for verifying why an action was permitted, where an error occurred, or how responsibility should be assigned. This paper presents the Structured Cognitive Loop as an architecture for accountable behavior in large language model agents. SCL separates cognition, memory, control, and action into distinct modules. The language model proposes. External memory preserves verified state. A lightweight controller checks preconditions, prevents redundant actions, and authorizes execution before tools are used. We evaluate SCL against ReAct and common LangChain agent variants across travel planning, conditional email drafting, and constraint guided image generation. Across 360 episodes, SCL achieves 86.3 percent task success compared with 70.5 to 76.8 percent for prompt based baselines. It also improves goal fidelity, reduces redundant tool calls, increases reuse of intermediate state, and lowers unsupported assertions. This extended revision situates SCL within a broader architecture of epistemic accountability. Subsequent extensions integrate context aware Human in the Loop control, Pool Gated Retrieval, and the Horizon Warrant Commitment framework. Together these components define an agent architecture in which the model proposes, structure decides, evidence is warranted before use, and human judgment is embedded in the trace rather than imposed after the fact. The result is a foundation for AI agents whose decisions are not only effective but also authorized, inspectable, and accountable.

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

MVM-IOD: An Industrial Object-Centric Benchmark Dataset for the Evaluation of 3D Reconstruction Methods

3D object reconstruction, and camera pose estimation in industrial applications are challenging tasks, as errors are costly while the computation time is often limited. The complexity of typical industrial objects further complicates these tasks. Most of the existing datasets in this context do not depict realistic industrial scenarios. Therefore, we introduce the Machine Vision Metrology Industrial Object Dataset (MVM-IOD). Images of typical industrial objects are captured systematically, by moving a camera, mounted at the end effector of an industrial robot arm, on a hemisphere around the objects. MVM-IOD contains reference camera poses and reference 3D point clouds, the acquired RGB images of 9 objects and 2 background choices resulting in 18 scenes, which allows evaluation of all image based methods that compute a 3D reconstruction, camera poses, or novel views of a scene. Based on MVM-IOD, we extensively evaluate current SOTA 3D reconstruction and camera pose estimation methods, such as Structure from Motion, Multi-View Stereo, recent feed forward methods (Visual Geometry Grounded Transformer, {\pi}3), and 2D Gaussian Splatting and report our findings as a baseline for future research. The experiments show that capture setups like ours generate out-of distribution images for feed forward methods, leading to suboptimal point clouds and camera poses. However, these out-of-distribution images can be shifted closer to the training distribution by applying simple preprocessing steps. Consequently, in certain industrial applications, feed forward methods should be used with caution.

20.
arXiv (CS.CL) 2026-06-17

EngTrace: A Symbolic Benchmark for Verifiable Process Supervision of Engineering Reasoning

Large Language Models (LLMs) are increasingly entering specialized, safety-critical engineering workflows governed by strict quantitative standards and immutable physical laws, making rigorous evaluation of their reasoning capabilities imperative. However, existing benchmarks such as MMLU, MATH, and HumanEval assess isolated cognitive skills, failing to capture the physically grounded reasoning central to engineering, where scientific principles, quantitative modeling, and practical constraints must converge. To enable verifiable process supervision in engineering, we introduce EngTrace, a symbolic benchmark built on 90 parameterized templates, each generating unique, contamination-resistant problem instances, spanning three major engineering branches, nine core domains, and 20 distinct areas, yielding 1,350 test cases that stress-test generalization across diverse physical scenarios. Moving beyond outcome matching, we introduce a verifiable two-stage evaluation framework that uses a tiered protocol to validate intermediate reasoning traces alongside final answers through automated procedural checks and a heterogeneous AI Tribunal. Our evaluation of 27 leading LLMs reveals a distinct trade-off between numeric precision and trace fidelity, identifying a complexity cliff where abstract mathematical pre-training fails to translate into the integrative reasoning required for advanced engineering tasks.

21.
arXiv (CS.CL) 2026-06-15

AdaSR: Adaptive Streaming Reasoning with Hierarchical Relative Policy Optimization

Large reasoning models typically follow a read-then-think paradigm: they observe the complete input, reason over a static context, and then produce the answer. Yet many real-world scenarios are inherently dynamic, such as audio and video stream, where information arrives as a continuous stream and models must reason, update, and respond under partial observations. Recent streaming reasoning methods allow models to think while reading, but they largely rely on supervised imitation of pre-constructed trajectories, which limits their flexibility. In this paper, we propose AdaSR, an adaptive streaming reasoning framework that enables models to reason during input streaming and perform final deliberation once the stream is complete, learning when to think, and how much computation to allocate across different stages. To optimize this hierarchical reasoning process, we introduce Hierarchical Relative Policy Optimization (HRPO), which decomposes policy optimization into streaming reasoning and deep reasoning phases, providing more fine-grained advantage assignment instead of uniformly distributing a single sequence-level advantage over all tokens. HRPO integrates format, accuracy, and adaptive thinking rewards to enforce valid reasoning protocols, preserve final task performance, and encourage latency-aware computation allocation. Experiments show that AdaSR achieves a better balance among reasoning accuracy, computational efficiency, and streaming latency compared with supervised fine-tuning baseline. We release our code at https://github.com/EIT-NLP/StreamingLLM/tree/main/AdaSR.

22.
arXiv (CS.CL) 2026-06-12

Demystifying Hidden-State Recurrence: Switchable Latent Reasoning with On-Policy Reinforcement Learning

Latent chain-of-thought compresses reasoning by replacing visible reasoning traces with continuous hidden-state recurrence, but existing formulations are difficult to optimize with standard on-policy reinforcement learning (RL) and hard to interpret causally. Our key insight is that a single pair of explicit boundary tokens can address both issues at once: discrete entry and exit anchors make the latent block compatible with standard on-policy RL, and the same anchors offer a natural foothold for mechanistic analysis. Motivated by this, we propose SWITCH, a switchable latent reasoning framework. The model emits to enter latent mode and to exit. Because the boundaries are ordinary discrete tokens, the GRPO policy ratio is well-defined at every decision point. The same anchors also expose the latent steps to direct probing and causal intervention. We train the model with a visible-to-latent curriculum and a Switch-GRPO objective that propagates gradients through recurrent latent computation. SWITCH consistently outperforms prior hidden-state-recurrence latent reasoning approaches at similar scale. Mechanistic analysis through the boundary tokens further reveals three findings: (i) is a sharply localised, learned switching policy rather than a stylistic artefact; (ii) the latent step it opens performs problem-specific, causally important computation rather than acting as an inert placeholder; and (iii) that computation is concentrated at a single hidden-state transition on entry. Together, these results show that hidden-state-recurrence latent reasoning is both RL-trainable and open to direct mechanistic analysis, including of how on-policy RL itself improves the model from the inside.

23.
arXiv (CS.CL) 2026-06-17

A Two-Phase Stability Study of LLM Judges and Bar Council Examiners on Thai Bar-Exam Free-Form Essays

Free-form legal essay evaluation in NLP treats expert inter-rater stability as a single ceiling number, and treats LLM-judge agreement with that ceiling as evidence of judge stability. We test both assumptions on the Thai bar examination through an identical-inputs protocol: three Bar Council-trained examiners (A, B, C) and a 26-LLM judge panel score the same 15 cross-graded answers from the same four inputs (question, official Bar Council grading regulation, gold answer, candidate answer). The headline finding is asymmetric. On 10 of 15 cells where the rubric prescribes both axes, all 29 raters converge in a tight band: panel agreement is universal. On the remaining 5 cells where the rubric does not prescribe how to grade a correct final answer that omits a decisive statutory citation, the human panel splits between two coherent readings (B/C majority at the upper rubric band, score 6-8; A minority at the lower band, score 1-2). The LLM judge population does not split symmetrically: 22 of 26 LLMs score in or near B/C's contested band, 3 sit in the regulation-silent middle gap, and only 1 (GPT-5.4 Nano) approaches A's band without consistently scoring within it. Zero LLMs in our 26-judge panel reproduce the minority human reading on the contested cells. The B/C-direction cluster spans every model size, vendor, and price tier we tested. An instrumented three-LLM anchor sub-panel (Claude 4.6 Opus, Gemini 3.1 Pro, GPT-5.4 Pro) carries determinism probes, input ablations, and bootstrap CIs, and reaches anchor panel $\alpha = 0.77$ on the 15 cells against human-panel $\alpha = 0.36$. The high LLM-panel $\alpha$ reflects systematic convergence on the majority reading rather than balanced reproduction of both readings; a benchmark that selects its LLM judge by maximising agreement with a human reference panel will inherit this asymmetry by construction.

24.
arXiv (CS.CV) 2026-06-18

CrossEarth-Gate: Fisher-Guided Adaptive Tuning Engine for Efficient Adaptation of Cross-Domain Remote Sensing Semantic Segmentation

In Remote Sensing (RS), Parameter-Efficient Fine-Tuning (PEFT) has emerged as a key approach to activate the generalizable representation ability of foundation models for downstream tasks. However, existing specialized PEFT methods often fail when applied to large-scale Earth observation tasks, as they are unable to fully handle the multifaceted and unpredictable domain gaps (e.g., spatial, semantic, and frequency shifts) inherent in RS data. To overcome this, we propose CrossEarth-Gate, which introduces two primary contributions. First, we establish a comprehensive RS module toolbox to address multifaceted domain gaps, comprising spatial, semantic, and frequency modules. Second, we develop a Fisher-guided adaptive selection mechanism that operates on this toolbox. This selection is guided by Fisher Information to quantify each module's importance by measuring its contribution to the task-specific gradient flow. It dynamically activates only the most critical modules at the appropriate layers, guiding the gradient flow to maximize adaptation effectiveness and efficiency. Comprehensive experiments validate the efficacy and generalizability of our method, where CrossEarth-Gate achieves state-of-the-art performance on 16 out of 18 cross-domain benchmarks for RS semantic segmentation.

25.
arXiv (CS.CL) 2026-06-19

LaViSA: A Language and Vision Structural Ambiguity Benchmark

Structural ambiguity arises when a single sentence admits multiple valid interpretations due to its syntactic structure, posing a fundamental challenge for language understanding. Visual scenes serve as useful cues for resolving such ambiguity, and Vision and Language Models (VLMs) need to be capable of deriving possible semantic interpretations from visual scenes. We introduce Language and Vision Structural Ambiguity (LaViSA), a benchmark designed to evaluate the ability of VLMs to resolve structural ambiguity leveraging visual scenes. LaViSA consists of ambiguous sentences, their disambiguated sentences, and corresponding images of these disambiguated sentences across seven ambiguity categories. Using LaViSA, we conduct a comprehensive evaluation of diverse VLMs, including both proprietary and open-source models with varying parameter scales and reasoning capabilities. Experimental results show that although recent VLMs can leverage visual scenes to resolve structural ambiguity to a some extent, they still struggle with certain ambiguity types and visually subtle semantic distinctions, indicating remaining limitations in resolving structural ambiguity using visual scenes.