Full author lists and current manuscript statuses are listed below. See also my Google Scholar profile.
† Equal contribution (co-first authors).
EMNLP 2026. Accepted.
Training-free, sampler-agnostic remasking guided by confidence drift improves accuracy across two diffusion LLMs, three samplers, and four benchmarks, with 76% fewer forward evaluations than ReMDM in a matched setting.
EMNLP 2026. Accepted.
A real-world data science agent benchmark with 641 tasks across 285 datasets, incorporating multimodal perception, multi-query interactions, and multidimensional evaluation of 13 advanced LLM agents.
IEEE Transactions on Artificial Intelligence. Accepted.
A Transformer-based autoencoder learns limit order book representations for multi-agent market calibration, achieving lower calibration errors than mid-price and raw-LOB approaches in 9 of 10 simulation scenarios.
Under review at ICLR 2027.
An information-theoretic connection between temporal token instability and masked-context dependency motivates stability-aware decoding. On LLaDA-8B, mean accuracy increases from 25.63% to 41.08% across four benchmarks while average forward evaluations decrease by 22.7%.
Under review at ICLR 2027.
A two-stage framework bootstraps a task-adapted teacher through reference-conditioned self-distillation, then trains an independent student via on-policy distillation. On Qwen3-4B, mean mathematical reasoning accuracy improves from 18.06% to 23.98%, surpassing the constructed teacher.
Under review at ICLR 2027.
Familiarity-adaptive token weighting based on pretrained-model familiarity improves mean mathematical reasoning accuracy from 21.32% to 37.20% over standard SFT on Qwen2.5-Math-7B.
Under review at ICLR 2027.
Privileged-information-guided token weighting uses a frozen, reference-conditioned model without external teachers or on-policy rollouts, improving mean mathematical reasoning accuracy from 36.41% to 47.80% over standard SFT on Qwen3-8B-Base.
Under review at the Journal of the American Statistical Association.
Distribution-aware retrieval integrates statistical data profiles into R package embeddings, achieving 93.47% NDCG@10 versus 79.32% for the strongest baseline and improving downstream agents on statistical analysis tasks.