Search papers, labs, and topics across Lattice.
This paper introduces the Exploratory-Assimilating Reflection (EAR) framework, designed to enhance memory retrieval in LLM-based autonomous agents by addressing issues of adaptability and sample efficiency. By integrating Exploratory Reflection for iterative search and Assimilating Reflection for experience replay, EAR significantly improves the retrieval of relevant memories from heterogeneous stores. Experimental results demonstrate that EAR achieves up to a 17.9% improvement in retrieval performance on long-term dialogue benchmarks, showcasing its robustness to noisy feedback.
EAR boosts memory retrieval performance by nearly 18% in long-term interactions, setting a new standard for LLM adaptability.
LLM-based autonomous agents require external memory to overcome their statelessness and limited context window for long-term interaction and dynamic knowledge reasoning. However, existing memory retrieval methods often lack adaptability and sample efficiency, and struggle to retrieve the right mixture of memories from heterogeneous stores. We propose Exploratory-Assimilating Reflection (EAR), a framework for high initial retrieval performance and sample-efficient adaptation. EAR combines two mechanisms: Exploratory Reflection, which performs iterative search to bootstrap retrieval and collect useful experiences for each query, and Assimilating Reflection, which replays these experiences from an Experience Buffer to refine a global reranker more efficiently than methods relying only on immediate rewards. Experiments show that EAR improves retrieval by up to 17.9% over the baseline retriever on two long-term dialogue benchmarks. We also show that EAR is highly sample-efficient and robust to noisy feedback.
