Hongyin Luo

A new model improves understanding of speech and sounds simultaneously.

ThReaD improves LLMs' performance on complex tasks through dynamic thread management.

Self-MoE creates specialized experts for improved language model performance.