Donatien Yeto Back to research

Research

MemHash: Associative Hash-Addressed Memory for AI Agents

Paper · In Progress

Investigating a novel memory architecture for AI agents combining semantic hashing, contextual associations, and sparse memory addressing. The research explores direct or near-direct access to associated memories beyond vector similarity search, including memories linked through experience or explicit relationships. Current research directions include sparse routing, knowledge-graph-augmented associations for structured knowledge representation, and scalable control of memory collision

Authors

  1. Donatien YETO

Topics

  • AI Agent Memory
  • Locality-Sensitive Hashing (LSH)
  • Knowledge Graphs - GraphRAG
  • Sparse Distributed Memory
  • Associative Memory
  • Approximate Nearest Neighbor (ANN)
  • Semantic Hashing