arXiv subjects

Yehan Yang

Publications and source records attributed to Yehan Yang.

7 recordsLinked to original sources

arXiv2026

Autonomy-of-Heads: Data-Free Sparse Attention from Frozen Query-Key Geometry

Long-context LLM inference is bottlenecked by quadratic attention computation and growing KV-cache costs. Existing sparse attention and KV-compression methods typically decide which tokens or heads to preserve from runtime attention scores, observation windows, calibration prompts, or learned gates, making head diagnosis input-dependent and costly to deploy. We propose Autonomy-of-Heads (AoH), a data-free method that identifies retrieval and streaming heads from the spectral geometry of query-key projections. AoH defines the kernel attention operator $M_h = W_K^{h\top}W_Q^h$ and uses its effective-rank as a weight-space measure of head function: concentrated spectra indicate a small number of dominant query-key matching directions and are associated with retrieval heads, whereas diffuse spectra indicate the absence of a dominant global matching direction and are associated with streaming heads. We further derive an efficient $d_\text{head}$-dimensional computation that avoids constructing the full $d_\text{model}\times d_\text{model}$ matrix. We conducted extensive experiments across models demonstrating that at 50\% sparsity, AoH retains 96.5\% of Full Attention performance on average while reducing prefill and decode latency by up to 41.4\% and 66.0\%, respectively, and KV-cache memory by 50.0\% at 256K tokens.

cs.CL↗

arXiv2026

Deconstructing Off-Policy Ratios: Entropy-Scaled Trust Regions for Asynchronous Reinforcement Learning

Asynchronous reinforcement learning (RL) accelerates large language model (LLM) post-training by overlapping rollout generation with policy optimization, but the resulting stale, off-policy data can destabilize optimization and ultimately cause policy collapse. Existing methods typically retain or discard tokens based solely on the magnitude of their importance ratios, applying the same threshold uniformly across token positions. In this work, we reveal that the natural scale of the importance ratio varies systematically with token entropy. Under asynchronous dynamics, this entropy-ratio scaling dictates two distinct phenomena: at low entropy, the inherent train-inference discrepancy is drastically amplified into substantial sampling noise; at high entropy, in-flight weight updates naturally induce pronounced, legitimate exploratory deviations. Consequently, magnitude-only correction inadvertently admits the amplified noise while strictly masking out the essential exploration triggered by in-flight updates. To address this, we propose the Entropy-Scaled Trust Region (ESTR), which scales each token's off-policy deviation by its local entropy, requiring no auxiliary forward passes or explicit version-switch detection. Across long-horizon agentic tasks and mathematical reasoning benchmarks, ESTR consistently outperforms existing asynchronous methods and achieves the best train-inference consistency. It reaches $37.34$ avg@1 on BrowseComp-Plus and $95.69$ on multi-turn GSM8K, matching synchronous GRPO while achieving a $2.6\times$ speedup.

cs.AI↗

arXiv2026

Beyond the Final Actor: Modeling the Dual Roles of Creator and Editor for Fine-Grained LLM-Generated Text Detection

The misuse of large language models (LLMs) requires precise detection of synthetic text. Existing works mainly follow binary or ternary classification settings, which can only distinguish pure human/LLM text or collaborative text at best. This remains insufficient for the nuanced regulation, as the LLM-polished human text and humanized LLM text often trigger different policy consequences. In this paper, we explore fine-grained LLM-generated text detection under a rigorous four-class setting. To handle such complexities, we propose RACE (Rhetorical Analysis for Creator-Editor Modeling), a fine-grained detection method that characterizes the distinct signatures of creator and editor. Specifically, RACE utilizes Rhetorical Structure Theory (RST) to construct a logic graph for the creator's foundation while extracting Elementary Discourse Unit (EDU)-level features for the editor's style. Experiments show that RACE outperforms 12 baselines in identifying fine-grained types with low false alarms, offering a policy-aligned solution for LLM regulation.

cs.CL↗

arXiv2025

PhantomHunter: Detecting Unseen Privately-Tuned LLM-Generated Text via Family-Aware Learning

With the popularity of large language models (LLMs), undesirable societal problems like misinformation production and academic misconduct have been more severe, making LLM-generated text detection now of unprecedented importance. Although existing methods have made remarkable progress, a new challenge posed by text from privately tuned LLMs remains underexplored. Users could easily possess private LLMs by fine-tuning an open-source one with private corpora, resulting in a significant performance drop of existing detectors in practice. To address this issue, we propose PhantomHunter, an LLM-generated text detector specialized for detecting text from unseen, privately-tuned LLMs. Its family-aware learning framework captures family-level traits shared across the base models and their derivatives, instead of memorizing individual characteristics. Experiments on data from LLaMA, Gemma, and Mistral families show its superiority over 7 baselines and 3 industrial services, with F1 scores of over 96%.

cs.CL↗

arXiv2025

From Judgment to Interference: Early Stopping LLM Harmful Outputs via Streaming Content Monitoring

Though safety alignment has been applied to most large language models (LLMs), LLM service providers generally deploy a subsequent moderation as the external safety guardrail in real-world products. Existing moderators mainly practice a conventional full detection, which determines the harmfulness based on the complete LLM output, causing high service latency. Recent works pay more attention to partial detection where moderators oversee the generation midway and early stop the output if harmfulness is detected, but they directly apply moderators trained with the full detection paradigm to incomplete outputs, introducing a training-inference gap that lowers the performance. In this paper, we explore how to form a data-and-model solution that natively supports partial detection. For the data, we construct FineHarm, a dataset consisting of 29K prompt-response pairs with fine-grained annotations to provide reasonable supervision for token-level training. Then, we propose the streaming content monitor, which is trained with dual supervision of response- and token-level labels and can follow the output stream of LLM to make a timely judgment of harmfulness. Experiments show that SCM gains 0.95+ in macro F1 score that is comparable to full detection, by only seeing the first 18% of tokens in responses on average. Moreover, the SCM can serve as a pseudo-harmfulness annotator for improving safety alignment and lead to a higher harmlessness score than DPO.

cs.CL↗

arXiv2025

FMASH: Advancing Traditional Chinese Medicine Formula Recommendation with Efficient Fusion of Multiscale Associations of Symptoms and Herbs

Traditional Chinese medicine (TCM) exhibits remarkable therapeutic efficacy in healthcare through patient-specific formulas. However, current AI-based TCM formula recommendation models and methods mainly focus on data-based textual associations between symptoms and herbs, and have not fully utilized their features and relations at different scales, especially at the molecular scale. To address these limitations, we propose the Fusion of Multiscale Associations of Symptoms and Herbs (FMASH), a novel framework that effectively incorporates the properties of herbs on different scales with clinical symptoms and provides refined embeddings of their multiscale associations. The framework integrates molecular-scale features and macroscopic properties of herbs and combines complex local and global relations in the heterogeneous graph of symptoms and herbs. Moreover, it provides effective representation embeddings of the multiscale features and associations of symptoms and herbs in a unified semantic space. Comprehensive experiments have been conducted on FMASH, and the results demonstrate that our FMASH-based model outperforms the state-of-the-art (SOTA) model on both datasets, confirming the effectiveness of FMASH in building the TCM formula recommendation model. In Dataset1, our model has achieved a significant improvement compared to the SOTA model, with increases of 3.38% in Precision@5, 3.89% in Recall@5, and 3.69% in F1-score@5. In Dataset2, Precision@5, Recall@5, and F1-score@5 increase by 2.64%, 1.92%, and 2.23%, respectively. This work facilitates the application of the AI-based TCM formula recommendation and promotes the innovative development of TCM diagnosis and treatment.

cs.LG↗

arXiv2025

Jingfang: An LLM-Based Multi-Agent System for Precise Medical Consultation and Syndrome Differentiation in Traditional Chinese Medicine

The practice of Traditional Chinese Medicine (TCM) requires profound expertise and extensive clinical experience. While Large Language Models (LLMs) offer significant potential in this domain, current TCM-oriented LLMs suffer two critical limitations: (1) a rigid consultation framework that fails to conduct comprehensive and patient-tailored interactions, often resulting in diagnostic inaccuracies; and (2) treatment recommendations generated without rigorous syndrome differentiation, which deviates from the core diagnostic and therapeutic principles of TCM. To address these issues, we develop \textbf{JingFang (JF)}, an advanced LLM-based multi-agent system for TCM that facilitates the implementation of AI-assisted TCM diagnosis and treatment. JF integrates various TCM Specialist Agents in accordance with authentic diagnostic and therapeutic scenarios of TCM, enabling personalized medical consultations, accurate syndrome differentiation and treatment recommendations. A \textbf{Multi-Agent Collaborative Consultation Mechanism (MACCM)} for TCM is constructed, where multiple Agents collaborate to emulate real-world TCM diagnostic workflows, enhancing the diagnostic ability of base LLMs to provide accurate and patient-tailored medical consultation. Moreover, we introduce a dedicated \textbf{Syndrome Differentiation Agent} fine-tuned on a preprocessed dataset, along with a designed \textbf{Dual-Stage Recovery Scheme (DSRS)} within the Treatment Agent, which together substantially improve the model's accuracy of syndrome differentiation and treatment. Comprehensive evaluations and experiments demonstrate JF's superior performance in medical consultation, and also show improvements of at least 124% and 21.1% in the precision of syndrome differentiation compared to existing TCM models and State of the Art (SOTA) LLMs, respectively.

cs.CL↗