Training-Free Lexical-Dense Fusion for Conversational-Memory Retrieval

Arxiv CS.CL (NLP/language models)

Proposes a training-free fusion of lexical and dense retrieval for long-term conversational memory. Addresses the retrieval bottleneck in tasks like LoCoMo and LongMemEval, building on concurrent work Nano-Memory.

See 2 more sources

Training-Free Lexical-Dense Fusion for Conversational-Memory Retrieval6 days agoChristian Lysenst{\o}en

LazyAttention: Efficient Retrieval-Augmented Generation with Deferred Positional Encoding7 days agoHaocheng Xia, Mihir Pamnani, Hanxi Fang, Supawit Chockchowwat, Yongjoo Park

Arxiv CS.CL (NLP/language models)·Rahul Subramani

7 days ago