arXiv subjects

Yidan Huang

Publications and source records attributed to Yidan Huang.

9 recordsLinked to original sources

arXiv2026

Authorization Drift in LLM Agents: Evaluating Boundary Compliance in Multi-Step Tool Use

LLM agents can exceed their authorization boundaries while pursuing user goals without an adversary. We formalize this phenomenon as authorization drift across data, tool, scope, and time, and introduce the Authorization Drift Benchmark (ADB): 648 samples across seven domains, each pairing a sufficient compliant path with an attractive unauthorized shortcut. To ensure executable tasks, tempting shortcuts, and feasible compliant paths, Staged Execution-Reflective Synthesis (SERS) grounds benchmark validation in real agent rollouts rather than LLM self-critique. Across 13 LLMs and three frameworks (25,207 trajectories), greater capability accompanies greater drift, with frontier-capable models exhibiting drift rates of 0.43 to 0.61. Yet 86.18% of tasks admit an observed clean-success trajectory, showing that drift is avoidable. Drift is persistent and dominated by early data over-collection; apparent improvements under agentic scaffolds coincide with collapsing clean-completion performance. To understand potential harms to individuals and society, we map drift points to legal provisions across four jurisdictions, connecting overreach to privacy and information-security concerns. Data minimisation is the most frequently implicated provision, and 93.5% of legally grounded drifting runs map to multiple jurisdictions, making the potential consequences of authorization drift more interpretable.

cs.CR↗

arXiv2026

ClueFlow: Learning Long-Horizon Search through Attributed Research-State Transitions

Long-horizon search requires agents to acquire relevant evidence and integrate it into an evolving research state. Existing methods maintain research states to guide exploration or assign turn-level progress signals, yet grounding these signals in explicit, attributable state changes remains underexplored. We introduce ClueFlow, which learns long-horizon search from attributed research-state transitions. ClueFlow constructs Source-State-Final attribution chains that connect evidence-acquiring browser actions, Evidence and Candidate updates, and the final answer supported by those updates. During search, ClueFlow records typed state changes and potentially supporting browser actions, then applies gold-conditioned semantic validation and source tracing after execution. We use this attribution to guide both supervised fine-tuning and reinforcement learning. During SFT, attribution selects which decisions receive supervision with their original interaction contexts. During Turn-PPO, the same attribution augments outcome-based turn advantages with credit assigned along the Source-State-Final chains. Transition-level validation also recovers useful supervision from gold-consistent state changes in trajectories with incorrect final answers. We evaluate ClueFlow across two Qwen backbones and four search benchmarks spanning live-web and fixed-corpus settings, observing consistent gains from SFT and further improvements from PPO. With Qwen3.5-35B-A3B, ClueFlow achieves 60.0% accuracy on BrowseComp and 72.0% on BrowseComp-Plus. In controlled BrowseComp-Plus experiments, attribution-guided SFT matches all-turn exact-context SFT using about 30% of the training tokens, while attribution-guided PPO outperforms outcome-only PPO by 4.1 pp. These results demonstrate that state-transition attribution enables efficient supervision and effective credit assignment for long-horizon search agents.

cs.AI↗

arXiv2026

Flexible Inverse Design of Optical Multilayer Coatings Using Discrete-Continuous Flow Matching

Optical multilayer coatings are used in imaging optics, lasers, telecommunications, solar cells, and radiative cooling, where the spectral response required by each application is achieved by a well-designed layer structure. However, existing design methods have limitations. Local methods refine a starting design efficiently but depend on its quality, and global searches require up to millions of transfer-matrix evaluations per target. Existing deep-learning inverse-design models are fast, but they are trained for a fixed set of materials and a fixed wavelength band, and they either optimize the thicknesses for a fixed material sequence or choose them from a discrete grid. In this paper, we present IrisFlow, a generative model that addresses these limitations by taking each candidate material as its optical constants and the target spectrum together with its wavelengths as input. It generates the material and the continuous thickness of every layer jointly, by coupled discrete and continuous flow matching in a single network. We demonstrate its performance through design with materials outside the training library and on bands beyond the training range, and through comparisons with classical optimizers and an existing deep-learning model. We also demonstrate its usability in a practical design workflow by designing four colored solar-reflective coatings with the calibrated optical constants of an existing deposition system. Combined with a local refinement of its ten best designs, it reduces the number of transfer-matrix evaluations about 500-fold relative to needle optimization at comparable accuracy. The code, trained model, and training and test data are publicly available.

physics.optics↗

arXiv2026

SearchAuditor: Auditing and Attributing Failures in Long-Horizon Search Agents

Deep search agents tackle challenging questions through long-horizon web interactions, a process that is both complex and fragile: small reasoning errors may propagate through long, noisy trajectories into fluent but incorrect answers. Diagnosing such failures is difficult, requiring the manual inspection of extremely long execution traces, which could be beyond human capacity. We therefore introduce SearchAuditBench, a benchmark that evaluates whether LLM auditors can localize, attribute, and repair these failures, thereby reducing the human burden. SearchAuditBench comprises 1,243 failed trajectories, averaging 73.1 messages and 65.1K tokens, collected from eight open-weight models on five deep-search benchmarks, each expert-annotated with the critical error step, a search-specific root cause, and a reference repair with grading rubrics. We further propose SearchAuditor, a multi-perspective auditing framework that effectively localizes, attributes, and repairs search-agent failures through evidence-grounded adjudication. Experimental results show that even the strongest baseline, when powered by a frontier model like GPT-5.5, attains only a 26.6% end-to-end pass rate. In contrast, our SearchAuditor consistently outperforms all baselines across different frontier models, achieving an end-to-end pass rate of 32.3%, and resuming failed runs with its repairs enables agents to better recover from errors.

cs.AI↗

arXiv2026

EvalVerse: Pipeline-Aware and Expert-Calibrated Benchmarking for Professional Cinematic Video Generation

The rapid evolution of generative video foundation models has propelled the field toward professional-grade cinematic synthesis. To achieve such demanding quality, the community transitions towards Reinforcement Learning (RL) and agentic workflows. However, reliable evaluation has emerged as a critical bottleneck. Existing benchmarks predominantly evaluate ''whether it is right'' (basic prompt-following) while fundamentally neglecting ''whether it is good'' (cinematic quality, acting, and aesthetics). Furthermore, current automated metrics lack the domain-specific rigor required to provide trustworthy signals, creating a severe credibility gap between human aesthetic perception and machine scoring. To bridge this gap, we introduce EvalVerse, a comprehensive, pipeline-aware, and expert-calibrated evaluation framework. We treat video generation assessment not merely as an engineering task, but as a core scientific problem: the systematic digitization of subjective cinematic expertise. First, we organize domain knowledge into an evaluation taxonomy aligned with the professional filmmaking workflow (pre-production, production, and post-production). Second, we distill human expert judgments into a curated dataset with large-scale human annotations. Third, we inject this knowledge into Vision-Language Models (VLMs) through an expert-calibrated fine-tuning strategy, enabling the VLM to perform explicit Chain-of-Thought reasoning. Compared to previous works, EvalVerse not only retains compatibility with foundational ''rightness'' metrics, but also significantly expands the criteria to ''goodness'' and broaden the task coverage to complex multi-shot sequencing and audio-visual integration. Consequently, by providing granular diagnostic signals, EvalVerse transcends a static leaderboard and establishes a fundamental infrastructure for future work, such as reward models and evaluator agent.

cs.CV↗

arXiv2026

JoyAI-LLM Flash: Advancing Mid-Scale LLMs with Token Efficiency

We introduce JoyAI-LLM Flash, an efficient Mixture-of-Experts (MoE) language model designed to redefine the trade-off between strong performance and token efficiency in the sub-50B parameter regime. JoyAI-LLM Flash is pretrained on a massive corpus of 20 trillion tokens and further optimized through a rigorous post-training pipeline, including supervised fine-tuning (SFT), Direct Preference Optimization (DPO), and large-scale reinforcement learning (RL) across diverse environments. To improve token efficiency, JoyAI-LLM Flash strategically balances \emph{thinking} and \emph{non-thinking} cognitive modes and introduces FiberPO, a novel RL algorithm inspired by fibration theory that decomposes trust-region maintenance into global and local components, providing unified multi-scale stability control for LLM policy optimization. To enhance architectural sparsity, the model comprises 48B total parameters while activating only 2.7B parameters per forward pass, achieving a substantially higher sparsity ratio than contemporary industry leading models of comparable scale. To further improve inference throughput, we adopt a joint training-inference co-design that incorporates dense Multi-Token Prediction (MTP) and Quantization-Aware Training (QAT). We release the checkpoints for both JoyAI-LLM-48B-A3B Base and its post-trained variants on Hugging Face to support the open-source community.

cs.CL↗

arXiv2026

Reimagining Legal Fact Verification with GenAI: Toward Effective Human-AI Collaboration

Fact verification is a critical yet underexplored component of non-litigation legal practice. While existing research has examined automation in legal workflow and human-AI collaboration in high-stakes domains, little is known about how GenAI can support fact verification, a task that demands prudent judgment and strict accountability. To address this, we conducted semi-structured interviews with 18 lawyers to understand their current verification practices, attitudes toward GenAI adoption, and expectations for future systems. We found that while lawyers use GenAI for low-risk tasks like drafting and language optimization, concerns over accuracy, confidentiality, and liability are currently limiting its adoption for fact verification. These concerns translate into core design requirements for AI systems that are trustworthy and accountable. Based on these, we contribute design insights for human-AI collaboration in legal fact verification, emphasizing the development of auditable systems that balance efficiency with professional judgment and uphold ethical and legal accountability in high-stakes practice.

cs.HC↗

arXiv2024

VLBI Astrometry of Radio Stars to Link Radio and Optical Celestial Reference Frames: Observing Strategies

The Gaia celestial reference frame (Gaia-CRF) will benefit from a close assessment with independent methods, such as Very Long Baseline Interferometry (VLBI) measurements of radio stars at bright magnitudes. However, obtaining full astrometric parameters for each radio star through VLBI measurements demands a significant amount of observation time. This study proposes an efficient observing strategy that acquires double-epoch VLBI positions to measure the positions and proper motions of radio stars at a reduced cost. The solution for CRF link compatible with individual VLBI position measurements is introduced, and the optimized observing epoch scheduling is discussed. Applying this solution to observational data yields results sensitive to sample increase or decrease, yet they remain consistently in line with the literature at the 1-sigma level. This suggests the potential for improvement with a larger sample size. Simulations for adding observations demonstrate the double-epoch strategy reduces CRF link parameter uncertainties by over 30% compared to the five-parameter strategy.

astro-ph.IM↗

arXiv2021

VOLKS2: a transient search and localization pipeline for VLBI observations

We present VOLKS2, the second release of "VLBI Observation for transient Localization Keen Searcher". The pipeline aims at transient search in regular VLBI observations as well as detection of single pulses from known sources in dedicated VLBI observations. The underlying method takes the idea of geodetic VLBI data processing, including fringe fitting to maximize the signal power and geodetic VLBI solving for localization. By filtering the candidate signals with multiple windows within a baseline and by cross matching with multiple baselines, RFIs are eliminated effectively. Unlike the station auto spectrum based method, RFI flagging is not required in the VOLKS2 pipeline. EVN observation (EL060) is carried out, so as to verify the pipeline's detection efficiency and localization accuracy in the whole FoV. The pipeline is parallelized with MPI and further accelerated with GPU, so as to exploit the hardware resources of modern GPU clusters. We can prove that, with proper optimization, VOLKS2 could achieve comparable performance as auto spectrum based pipelines. All the code and documents are publicly available, in the hope that our pipeline is useful for radio transient studies.

astro-ph.IM↗