Searcharxiv⌕ Search

arXiv subjects

Liang Chen

Publications and source records attributed to Liang Chen.

At least 19 recordsLinked to original sources
arXiv2026

ScienceVision: Can AI Observe and Reason Like a Scientist?

Scientific discovery builds on accurate observations interpreted through domain knowledge and reasoning. For scientific 3D data, this requires understanding structures and spatial relationships, from protein contacts to anatomical organization. Existing multimodal benchmarks largely evaluate text and 2D images, while scientific 3D evaluations vary in representations and protocols across disciplines, limiting cross-disciplinary comparison. To assess these foundational capabilities, we introduce ScienceVision, a cross-domain benchmark for scientific 3D understanding and reasoning through expert-level visual question answering (VQA). We construct 250 questions, with 50 each from Biology, Chemistry, Medical Imaging, Mechanical CAD, and Architecture, through a pipeline combining synthetic question generation with domain-expert review and annotation. Each task pairs a native 3D artifact with a scientific question, a specified response format, and a task-specific reference answer, requiring domain knowledge to be applied to geometric or topological evidence. We evaluate 12 models on the same questions in three settings: visual question answering with native 3D artifacts accessed through an agent harness's file and code-execution tools; the same workflow augmented with expert-inspired viewer inspection and measurement tools exposed through the Model Context Protocol (MCP); and visual question answering from prescribed rendered 2D images using an agent harness. All settings use the same task-specific deterministic final-answer scoring. The inspection environment supports view changes, target isolation, and slice inspection. In this main comparison, the best mean answer score is 0.53 out of 1. Mean scores in the native-artifact workflow without additional MCP tools exceed those in the rendered-image workflow for every model, by 17.2 percentage points on average.

cs.CV↗
arXiv2026

NAS-PO: Native Attention-Strategy Policy Optimization for Vision-Language Models

Reinforcement learning with verifiable rewards (RLVR) has become an effective post-training paradigm for vision-language models (VLMs). Standard RLVR, however, assigns identical outcome-derived credit to all generated tokens regardless of their perceptual relevance. Recent perception-aware methods estimate visual reliance through image perturbation or masking. These indirect proxies require perturbation branches, while corruption sensitivity does not necessarily identify the visual evidence used for reasoning. We instead shape model perception through native attention, an intrinsic model signal. Rather than treating attention as a post-hoc explanation, we formulate it as a reward-anchored, token-aligned routing signal that encourages attention patterns supporting visually grounded reasoning. To this end, we propose NAS-PO, a Native Attention-Strategy Policy Optimization framework. NAS-PO analyzes native attention with a visual mass policy that quantifies visual allocation and a conditional visual policy that captures spatial routing over visual tokens, making the optimization well-posed. Based on this, Native Attention-Strategy Advantage Shaping (NAS-AS) draws on the visual-mass policy to redistribute signed policy-gradient credit for visually engaged rollouts. The Adaptive Attention-Coupled (AAC) loss then aligns the visual-mass policy on all response tokens and the conditional visual policy on visually active ones, consolidating the visual-routing patterns that drive successful trajectories. Experiments demonstrate that NAS-PO substantially surpasses leading RL-tuned VLMs across 7 out-of-domain reasoning and perception benchmarks, with its effectiveness consistently validated across 8B and 32B. Visual-Routing Sensitivity (VRS) analysis further indicates that the optimized attention prioritizes visually load-bearing regions.

cs.CV↗
arXiv2026

Step-wise Rubric Rewards for LLM Reasoning

Reinforcement Learning with Verifiable Rewards (RLVR) improves language-model reasoning using final-answer correctness, but lacks supervision for intermediate steps. Rubric-based methods such as Rubrics as Rewards (RaR) provide finer-grained criteria, yet collapse rubric scores into a response-level scalar. This loses criterion-level structure, rewards flawed steps in correct responses, penalizes valid steps in incorrect responses, and can encourage unbounded self-correction. In 1,000 sampled problems, 18.2% of steps in answer-correct responses are wrong, while 49.9% of steps in answer-incorrect responses are correct. We propose Step-wise Rubrics as Rewards (SRaR), which (i) uses an LLM judge to assign rubric items to reasoning steps, (ii) normalizes step scores across rollouts to focus learning on quality differences, and (iii) combines step-wise and outcome rewards through a decoupled advantage estimator that preserves a stable outcome baseline. We also construct a 16K-problem rubric dataset by contrastively distilling criteria from verified correct and flawed reasoning paths. Across six math benchmarks, SRaR improves average accuracy over RaR by 3.57 and 2.75 points on Qwen3-8B-Non-Thinking and Qwen3-32B-Non-Thinking, respectively. On AIME 2025, it increases the Faithful Reasoning Rate from 34.5% to 46.7% and reduces self-correction looping from 48.1% to 26.5%. SRaR also achieves the best results on three of four code-generation benchmarks.

cs.LG↗
arXiv2026

Anchor 1: Precise and Stable Humanoid Control for Whole-Body Manipulation

Humanoid controllers have acquired increasingly diverse motor skills, yet translating these capabilities into precise and stable whole-body manipulation remains challenging. Manipulation tasks are sensitive to execution deviations -- even small errors or unintended steps can cause failure. Introducing Anchor 1, a whole-body controller for precise and stable whole-body manipulation, together with a benchmark spanning diverse coordination demands. Anchor 1 adopts calibrated sparse motion references to preserve task-relevant human motion structure without online joint-space retargeting, leaving kinematic freedom for precise control. To guide this freedom towards stable, physically plausible whole-body coordination, we shape the training distribution through offline curation of robot motions paired with these calibrated sparse references. We further introduce explicit error feedback decoupled from motion encoding, improving control precision while supporting flexible command interfaces. To achieve a systematic evaluation of these capabilities, a benchmark is proposed to assess precision, sustained execution stability, and reachability across diverse whole-body coordination demands. Evaluations in simulation and on hardware show stronger precision and stability across all evaluation tracks. Under the most strict real-world global scoring setting, Anchor 1 achieves a 30\% to 80\% decrease in tracking error with fewer look-ahead inputs, and without retargeting compared to all existing controllers. Downstream experiments demonstrate more efficient teleoperated data collection and higher task success using the same deployment pipeline as well. We believe Anchor 1 supports precise and stable whole-body control as a practical basis for humanoid task execution.

cs.RO↗
arXiv2026

PAPERBENCHX: Can AI Reproduce Science Across Disciplines?

Before AI can be trusted to discover new science, it should be able to reproduce what science already knows. Reproduction provides a rigorous checkpoint for an AI scientist: the target claim is known, yet recovering it still requires translating an incomplete scientific description into an executable experiment and producing evidence that can be independently verified. Existing research-reproduction benchmarks largely focus on AI and machine learning, where experiments share relatively homogeneous software stacks. Extending this capability across scientific disciplines introduces a different challenge: agents must operate domain-specific scientific software, make physical and numerical choices, diagnose long-running computations, and distinguish valid evidence from merely plausible outputs. We introduce PAPERBENCHX, a benchmark of 93 expert-constructed paper-reproduction tasks spanning 12 research areas across electromagnetics, photonics, chemistry, materials science, biology, and robotics. Each task provides the source paper, a public task specification, and a domain-native scientific environment involving specialized scientific software. Agents must reconstruct a key experiment, execute and diagnose it, and deliver a runnable workflow that supports the target scientific claim. To evaluate scientific validity rather than surface-level agreement, we replay every submission from scratch and score only regenerated evidence against expert-validated criteria. Across 10 frontier agent configurations, the best configuration fully reproduces only 14% tasks under the 90% threshold. Meanwhile, mean rubric scores range from 27.53% to 59.08%, indicating that substantial partial progress rarely translates into complete scientific reproduction. These results reveal a large gap between paper understanding and reliably producing reproducible scientific evidence.

cs.AI↗
arXiv2026

A Candidate Binary Microlensing Event in the 2018 Gamma-Ray Flare of 3C 279

The April 2018 gamma-ray flare of the flat-spectrum radio quasar 3C~279 exhibited sharp peaks within a broader outburst. Its rapid substructure was previously attributed to magnetic reconnection. Motivated by its distinctive temporal profile within our comparison sample of bright blazar flares and its resemblance to caustic-crossing features in optical microlensing events, we investigate a finite-source binary-microlensing interpretation. A jointly fitted model of binary-lens magnification, a smooth intrinsic flare, and Gaussian-process variability reproduces the principal double-peaked structure of the GeV light curve. The adopted local solution has a binary mass ratio of $q\simeq0.098$ and an Einstein timescale of $t_{\rm E}\simeq0.327$~d. The absolute lens mass and projected source radius depend on the lens distance and relative proper motion. The fitted geometry can correspond to a Galactic planetary-mass binary or an intervening stellar or substellar system, with different physical source sizes in each case. We present binary microlensing as a testable interpretation of this event, with the finite-source structure offering a potential probe of the compact GeV-emitting region.

astro-ph.HE↗
arXiv2026

Numerical Simulations of the interaction between winds from tidal disruption event and supermassive black hole circumnuclear medium

The interaction between winds from tidal disruption event (TDE) and the circumnuclear medium (CNM) around supermassive black holes is believed to contribute to the radio emission of non-jetted TDEs. This model has been studied both analytically and numerically. In previous works, the properties of wind are usually set to be angle-independent. It is not clear to what extent the results deviate from those in the case where TDE wind properties are angle-dependent. We perform numerical simulations to investigate the effects of the angular distribution of wind properties on the radio light curve in the wind-CNM interaction model. We find that given that the wind mass and kinetic energy fixed, the radio light curve in the optically thin waveband significantly depends on the angle-dependence of wind properties. Even when the swept-up CNM mass is still less than the wind mass, a slight deceleration can affect the light curve. Changing the angular dependence of the wind properties causes the deceleration to vary across angles. Consequently, the final light curve--the time evolution of radiation integrated over all angles--would differ significantly from that predicted in models with a spatially uniform wind injection. This can result in a slower rise or a faster decline (depending on the CNM profile) than the analytical prediction, and can even turn the rise phase into a long-lasting plateau. Therefore, it is essential to carefully take into account the angular dependence of wind properties to study the wind-CNM interaction.

astro-ph.HE↗
arXiv2026

Composition factors and finite representation type

We prove that an Artin algebra has finite representation type if its indecomposable finitely generated modules are determined, up to isomorphism, by their composition factors. This gives an affirmative answer to the conjecture on composition factors and finite representation type recorded by Auslander, Reiten and Smalo.

math.RT↗
arXiv2026

Convergence Analysis of Variable-Separation-Based Model Reduction for Linear Parametric Partial Differential Equations

This paper analyzes the convergence of a variable-separation (VS) model reduction method that successively constructs separated approximations to the solutions of linear parametric partial differential equations (PDEs) by solving residual problems at sampled parameter values. We prove that the resulting VS approximations converge almost surely to the exact parametric solution in the parameter-averaged energy norm, under finite-dimensional approximation conditions on the error family and the assumptions of symmetry, uniform coercivity, and uniform boundedness of the governing bilinear forms. For parametric PDEs whose operators and right-hand sides admit finite-term separated representations, we construct finite-dimensional approximation spaces for the error family and bound their dimensions and approximation errors, thereby verifying these approximation conditions and obtaining an $O((\log N)^{-1})$ upper bound for the expected parameter-averaged error energy after $N$ iterations. When the solution families are contained in a fixed finite-dimensional space, we obtain the sharper $O(N^{-1})$ bound for the expected error energy and, if the operator is also parameter-independent, exact recovery for almost every parameter after an almost surely finite number of iterations. Numerical experiments illustrate the error decay of the VS approximations, as well as finite-step recovery in the parameter-independent case.

math.NA↗
arXiv2026

LoopGR: Parameter-Efficient Scaling for Generative Recommendation via Recurrent Depth

Generative recommendation formulates item recommendation as autoregressive generation rather than a discriminative scoring problem. Although empirical studies exhibit favorable scaling behavior in recommendation, unleashing the benefits of scaling remains challenging under the strict model size and inference cost constraints of real-world systems. A central challenge is to combine expressive modeling of user-item interaction histories with flexible computational depth without proportionally increasing model size. We introduce LoopGR, a parameter-efficient scaling framework for generative recommendation via recurrent depth. LoopGR employs a tailored hierarchical recurrent depth mechanism that alternates between a shared generative decoder and a lightweight context updater. For efficient inference, we further develop recurrent trajectory self-distillation that distills a low-budget student from a high-budget teacher by aligning both predictive distributions and evolution directions. Extensive experiments show that LoopGR consistently outperforms strong generative recommendation baselines. Compared with a representative encoder-decoder baseline, LoopGR achieves superior performance while reducing model size by up to 52%. These results establish recurrent depth as a new scaling dimension, offering a flexible and resource-efficient scaling path for generative recommendation.

cs.IR↗
arXiv2026

Search for Ultra-Relativistic Magnetic Monopoles with LHAASO-KM2A

Ultra-relativistic magnetic monopoles traversing the atmosphere are expected to initiate extensive air showers characterized by a muon deficit and a compact lateral profile. Using 1438.81 live days of data from the LHAASO-KM2A array, we present a search for these cosmic particles within the Lorentz factor range of $10^5$ to $10^7$. A machine-learning classifier utilizing lateral shower shape observables from the electromagnetic detectors is deployed to separate the signal from the hadronic background. No statistical excess above the background expectation is observed, and upper limits at the 90\% confidence level are established on the isotropic monopole flux. The resulting upper limit is $2.6 \times 10^{-16}\,\mathrm{cm}^{-2}\,\mathrm{s}^{-1}\,\mathrm{sr}^{-1}$ at $γ= 10^5$ and reaches $8.8 \times 10^{-19}\,\mathrm{cm}^{-2}\,\mathrm{s}^{-1}\,\mathrm{sr}^{-1}$ at $γ= 10^7$, which improves upon the astrophysical Parker bound by over three orders of magnitude. This work establishes constraints on the monopole flux within a previously unconstrained velocity window.

astro-ph.HE↗
arXiv2026

Local Allow, Global Deny: Why Per-Action Authorization Does Not Compose in LLM Agents

Autonomous LLM agents act through tools under per-action authorization: a local guard screens each tool call in isolation. We show that local authorization does not compose into global authorization. A compositional authorization failure (CAF) is a locally authorized execution, or a pair of locally authorized executions, that violates a global policy spanning intent, purpose, provenance, consent, and delegation. We characterize CAFs along two axes, expressiveness (single-trace safety versus relational k-safety) and observability (a ladder of monitor profiles). Our central result is an impossibility: relational purpose violations form a 2-safety hyperproperty with no per-trace locus of blame, so a per-action monitor that never raises a false alarm detects none of them, independent of the underlying model or decision procedure; they become visible only under cross-execution monitoring, which none of the six guardrails we survey implements. We make this constructive with AL-SYNTH, a SAT/SMT synthesizer (using no language model) that emits minimal CAF witnesses, each checked against an exhaustive oracle. The empirical signature is sharp: across nine LLM judges the per-action view approves the synthesized witness almost without exception, blocking it at a rate comparable to a benign control, yet the same models flag it when shown the two executions together. Per-action authorization is therefore structurally insufficient for relational policy conjuncts not reducible to single-trace safety: to bound them, a monitor must compare executions or accept false alarms.

cs.CR↗
arXiv2026

ACE-Compression: An Ability-Centric Evaluation of Compressed Large Language Models

Model compression makes large language models deployable, but today's evaluation tells an incomplete story. Standard practice measures retained task performance and stops there. This misses a critical fact: compression perturbs the distributed knowledge embedded across billions of weights, and task scores alone cannot reveal which underlying abilities survive and which quietly collapse. To solve this question, we introduce ACE-Compression, an ability-centric evaluation framework. Instead of asking only "does the compressed model still answer correctly?", we ask "which of its core abilities remain intact?" Our framework probes four deployment-facing dimensions: procedural reasoning and constraint execution, knowledge grounding and truthfulness, social cognition and affective understanding, and persona fidelity and stylistic consistency. Across more than 35 compressed checkpoints from six open-weight LLMs-covering quantization and pruning-we find that compression redistributes ability profiles. Coarse, local, and surface-level signals often persist, while fine-grained, long-tail, and context-dependent abilities become fragile. Quantization consistently outperforms pruning in retention, but the practical meaning of that gap depends entirely on which ability an application requires. Based on our findings, deployers can choose compression methods by the ability profile their application needs, not just by a leaderboard average. And algorithm designers gain diagnostic signals to build compression methods that preserve what matters. We believe that ACE-Compression turns compression from a black-box accuracy drop into an interpretable intervention-one we can measure, understand, and ultimately improve.

cs.AI↗
arXiv2026

On the Nonoccurrence of Gradient Restart in Nesterov Accelerated Gradient Method

Gradient restarting often improves the practical performance of accelerated gradient methods for unconstrained convex optimization problems, but a general theoretical guarantee of such improvement remains unclear. In this paper, we construct smooth, strongly convex objectives with Lipschitz-continuous gradients and corresponding initial points for which gradient restart never occurs, both for Nesterov's accelerated gradient method and for its ordinary differential equation model. From these examples, we conclude that strong convexity and smoothness alone do not guarantee that gradient restart improves the convergence behavior of the accelerated method or its continuous-time model. In particular, for the continuous-time example, the objective gap is asymptotic to $1/(πt^3)$. We therefore obtain a negative answer to the open problem raised by Su, Boyd, and Cand{è}s [\textit{J. Mach. Learn. Res.}, 2016, 17(153), 1--43] of whether gradient restart guarantees linear convergence of the continuous dynamics for smooth strongly convex objectives.

math.OC↗
arXiv2026

BridgeAfford: Bridging Query-Target Gap for Generalizable 3D Affordance Grounding

Language-guided 3D affordance grounding aims to localize functional regions on 3D objects according to interaction intents. However, existing methods exhibit poor generalization to unseen objects. One challenge is the query-target gap: a semantic discrepancy between query features encoding interaction intents and target-region features, leading to inaccurate grounding. End-to-end training favors query representations tailored to seen objects, limiting their compatibility with shifted target-region features on unseen objects. To address this issue, we propose BridgeAfford, a framework that reduces the query-target semantic gap from two complementary perspectives. First, Interaction Region Alignment (IRA) promotes cross-object alignment among geometrically similar interaction regions during training. This enables structures on unseen objects that resemble previously observed ones to produce representations similar to those learned during training, facilitating target-region recognition. Second, Visual Evidence Aggregation (VEA) leverages a 2D affordance model to refine the attention distribution and guide the query toward the correct region, yielding more target-aware queries. Extensive experiments on the LASO and PIAD datasets demonstrate that BridgeAfford outperforms existing methods, particularly on unseen objects, highlighting its robust generalization ability.

cs.CV↗
arXiv2026

DoubtLess: Training-Free Contrastive Decoding for Efficient Reasoning

Reasoning models achieve strong performance through long chains of thought (CoT), but often spend more tokens than needed to solve a problem. Existing ways to shorten the CoT require additional training for each model, add a fixed steering vector to the model's hidden states, or use decoding heuristics such as stopping early or banning words like "wait", which can reduce accuracy. We propose DoubtLess, a training-free decoding method in which a much smaller reasoning model steers the large one towards brevity. The small model reads the large model's CoT so far twice, once prompted to be brief (the brief thinker) and once prompted to be verbose (the verbose thinker), and the difference between its two logit vectors, scaled by a strength, is added to the large model's logits at every step of the CoT. On seven benchmarks of competition mathematics, graduate-level science and competitive programming, DoubtLess shortens the CoT of Qwen3-32B by 31% with unchanged accuracy, outperforming five inference-time baselines. We also show that DoubtLess reduces thinking tokens per turn by 20% in a self-debugging coding agent with unchanged task performance. It also generalizes to two other model families with unchanged accuracy. When combined with training-based compression, it further shortens the CoT of a model already trained for brevity. Ablations and behavioral analysis show that DoubtLess reduces repeated checking, with most token savings coming after the correct answer first appears in the CoT. Our code is available at https://github.com/leoluolol/DoubtLess.

cs.CL↗
arXiv2026

CivWorldBench: Evaluating Language World Models in Structured Multi-Agent Sandbox Worlds

The growing use of language models for forecasting motivates a broader evaluation of their world-model capabilities. Such evaluation must go beyond predictive accuracy to assess hidden-state inference and intervention prediction. Real-world forecasting benchmarks offer limited access to hidden-state ground truth and controlled experiments. We introduce CivWorldBench, a benchmark built on a multi-agent strategy game engine with recorded internal states, controlled observations, and paired intervention rollouts. Its 1,900 decision points from 475 worlds support three task levels: association forecasting, belief inference, and intervention prediction. Reference predictors support attributable evaluation of capability gaps, with an expert harness and empirical ceiling estimates contextualizing scores. Stored trajectories make intermediate predictions verifiable, while agent-play experiments test whether world-model capabilities support decision-making. The main panel comprises eleven frontier language models. Composite benchmark scores correlate with real-world forecasting leaderboards on matched model panels, more closely than with a general-capability index, supporting its use as a proxy for real-world forecasting capability. Strong association forecasts do not consistently extend to belief inference or intervention prediction, and effect magnitudes remain poorly calibrated despite informative direction predictions. Process verification reveals errors missed by final-answer evaluation, and structured reasoning improves intervention prediction for the strongest model in the process study. Agent-play experiments identify useful assisted plans and motivate evaluating decision-relevant capabilities beyond association forecasting. These findings identify hidden-state inference, intermediate-state prediction, and effect calibration as priorities for world-model training.

cs.CL↗