Jun Wu

Grounded in robust evidence retrieval, this workflow ensures customer-service LLMs adhere to policy rules while significantly enhancing answer quality.

SpikeTimer achieves a remarkable balance between copyright protection and performance, maintaining high accuracy on authorized data while effectively misclassifying unauthorized inputs.

Robots can now navigate complex environments without continuous goal updates, relying solely on their internal spatial memory.

Label-Specific Distance-based Oversampling reveals that tailoring synthetic instance generation to label-specific feature relevance can drastically enhance multi-label classification performance.

Achieving high accuracy in multi-speaker transcription, SoulX-Transcriber outperforms existing models by effectively addressing speaker overlap and rapid turn-taking.

Graph RAG systems leak almost all of their knowledge graph structure to black-box queries, even with existing safeguards.