Reproduce “Towards a Formal Definition of Agent Memory: Basis, Span, Optimality, and the Sequential Memory Problem” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Hongyao Tang

Researchers at Tianjin University introduced a formal framework for agent memory, defining memory as a basis for generated knowledge and establishing an optimality theory based on a utility-capacity frontier. This framework also formalizes memory management as a sequential Markov Decision Process and distinguishes between coverage and precision in noisy environments.

Key insights

  • Agent memory can be rigorously conceptualized as a basis-span relationship, where knowledge generation from events defines answerability as a coverage problem.
  • The 'utility as a universal yardstick for memory system efficiency, revealing that optimal memory design under constraints is generally an NP-hard problem.
  • Effective memory systems must account for noisy information by distinguishing between coverage and precision utility, and their dynamic management over time can be modeled as an online learning problem via an MDP.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.