Zhongle Xie

LLM embedding models waste over 40% of their inference compute processing prefix tokens that become completely redundant at deeper layers, enabling aggressive, training-free token pruning with virtually zero retrieval degradation.

Integrating AI into databases introduces a complex web of challenges, from query optimization to security, that demand a fundamental rethinking of traditional database architectures.

A new unified LVLM, OmniCT, bridges the gap between slice-level detail and volumetric understanding in CT scans, outperforming existing methods by a significant margin.

LLMs can be adapted for scenario-specific user representation by conditioning on queries, achieving state-of-the-art performance on Alipay benchmarks and demonstrating practical effectiveness in online A/B testing.