Jun Wang

AIR achieves a staggering 400x inference speedup for cross-domain recommendations without sacrificing semantic quality, revolutionizing industrial applications of LLMs.

LatentSkill achieves a 21.4-point increase in task success while slashing prefill token usage by over 64%, revolutionizing how LLM agents utilize skills.

Forget symmetrical fusion: GS-Fuse selectively incorporates event text into financial forecasting only when it demonstrably improves predictions beyond what historical prices alone can achieve.