Enhancing Long-Term Memory using Hierarchical Aggregate Tree for Retrieval Augmented Generation
This addresses the challenge of enabling more consistent, grounded long-form conversations from LLMs, though it appears incremental as it builds on existing retrieval-augmented generation techniques.
The paper tackles the problem of limited context capacity in large language models for long conversations by proposing a Hierarchical Aggregate Tree memory structure, which improves dialog coherence and summary quality over baseline contexts.
Large language models have limited context capacity, hindering reasoning over long conversations. We propose the Hierarchical Aggregate Tree memory structure to recursively aggregate relevant dialogue context through conditional tree traversals. HAT encapsulates information from children nodes, enabling broad coverage with depth control. We formulate finding best context as optimal tree traversal. Experiments show HAT improves dialog coherence and summary quality over baseline contexts, demonstrating the techniques effectiveness for multi turn reasoning without exponential parameter growth. This memory augmentation enables more consistent, grounded longform conversations from LLMs