Ling Luo

LLMs can now reliably extract complex, n-ary drug combinations from biomedical text, surpassing previous methods that were limited to binary interactions.

RAGNav's dual-memory system lets agents navigate complex, multi-goal environments without getting lost in spatial hallucinations, outperforming standard RAG methods.

Overcome cognitive overload in long-horizon VLN with a master-slave multi-agent architecture that distributes tasks and dynamically optimizes via dual-level reflection.

Softmax outputs hamstring conformal prediction's ability to adapt to input difficulty; using Helmholtz Free Energy from the logit space as a nonconformity score re-weighting mechanism unlocks more efficient and adaptive prediction sets.