MS student, Harbin Institute of Technology
3 papers at NeurIPS 2025
We introduce PHYSICS, a dataset containing 16,568 high-quality physics problems spanning subjects and difficulty levels, and an improved evaluation framework combining rule- and model-based judgments to advance LLMs' physical reasoning.
We propose an information-preserving and data-free pipeline for ultra-efficient delta compression, achieving ultra-high compression ratios and strong performance across various models.
Self‑supervised MoE‑Gyro restores clipped gyro peaks and denoises without labels, boosting range ±450 deg/s MEMS gyroscopes to ±1500 deg/s and cutting bias instability by 98 %.