arXiv subjects

Tianlei Chen

Publications and source records attributed to Tianlei Chen.

3 recordsLinked to original sources

arXiv2026

Counteraction-Aware Multi-Teacher On-Policy Distillation for General Capability Recovery with Domain Preservation

Domain specialization can improve LLM behavior, but often weakens the general capabilities inherited from the original model. Recent Multi-Teacher On-Policy Distillation (MOPD) pipelines recover model capabilities by supervising student-generated trajectories with teacher feedback, but typically assume teacher-aligned prompt coverage, requiring prompts to match the teachers' training distributions. This assumption is difficult to satisfy when the general teacher is an open-source model whose post-training data are unknown. Instead of attempting to reconstruct this hidden distribution, we study general capability recovery with readily available proxy general prompts. We identify two failure modes of vanilla MOPD in this incomplete-coverage situation: recovery-preservation counteraction from mixing conflicting recovery and preservation gradients, and weak-signal flattening from uniformly averaging samples with unequal correction demand. We propose \textbf{Counteraction-Aware Multi-Teacher On-Policy Distillation} (\textbf{CaMOPD}), which addresses these issues with decoupled alternating training and gap-based sample selection. CaMOPD allocates dedicated updates to general recovery, periodically performs domain-preservation updates, and selects samples with larger averaged token-level teacher-student log-probability gaps to concentrate correction signals. Across role-play dialogue and medical reasoning QA scenarios, CaMOPD outperforms all baselines in general capability recovery while maintaining domain-specific behavior. Gradient coherence analyses further support the intended effect of CaMOPD in producing more coherent correction signals.

cs.AI↗

arXiv2026

The Curse of Helpfulness: Inverse Scaling Law in Robustness to Distractor Instructions via DistractionIF

Large Language Models (LLMs) are increasingly deployed in agentic and retrieval-augmented generation (RAG) systems, where they must execute user-specified tasks over externally provided reference text. In practice, such context is often unstructured and contaminated with benign but instruction-like semantic noise, such as editorial comments and system traces, which should be treated strictly as data. We introduce DistractionIF, a benchmark designed to evaluate robustness against such distractor instructions in reference text. Across a broad range of models, we observe a consistent inverse scaling phenomenon: larger models are often less robust, with performance dropping by up to 30 points as scale increases. Mechanistically, our perplexity analysis reveals that scaling erodes the probabilistic boundary between robust and distracted behaviors, making models increasingly prone to over-interpreting noise as instructions. To address this, we demonstrate that reinforcement learning, specifically Group Relative Policy Optimization (GRPO), can restore this boundary, improving robustness by up to 15.5% without compromising general instruction-following capability. Our findings highlight a critical instruction-following robustness gap in reference-grounded tasks and establish reinforcement learning as a promising path for enforcing strict data-instruction separation at scale.

cs.AI↗

arXiv2026

Enhancing Rare Codes via Probability-Biased Directed Graph Attention for Long-Tail ICD Coding

Automated international classification of diseases (ICD) coding aims to assign multiple disease codes to clinical documents and plays a critical role in healthcare informatics. However, its performance is hindered by the extreme long-tail distribution of the ICD ontology, where a few common codes dominate while thousands of rare codes have very few examples. To address this issue, we propose a Probability-Biased Directed Graph Attention model (ProBias) that partitions codes into common and rare sets and allows information to flow only from common to rare codes. Edge weights are determined by conditional co-occurrence probabilities, which guide the attention mechanism to enrich rare-code representations with clinically related signals. To provide higher-quality semantic representations as model inputs, we further employ large language models to generate enriched textual descriptions for ICD codes, offering external clinical context that complements statistical co-occurrence signals. Applied to automated ICD coding, our approach significantly improves the representation and prediction of rare codes, achieving state-of-the-art performance on three benchmark datasets. In particular, we observe substantial gains in macro-averaged F1 score, a key metric for long-tail classification.

cs.LG↗