Tag
This paper presents a holistic evaluation of memory substrates for memory-augmented LLM agents, finding that no single substrate dominates across all regimes and advocating for adaptive substrate routing to optimize performance in different operating conditions.
This paper introduces dependency-guided rollback repair for memory-augmented agents, a method that builds a typed memory-to-action graph from runtime provenance to selectively undo faulty memory effects while preserving benign state, achieving strong recovery on benchmarks.
MemHarness is a framework that enables LLM agents to reconstruct past experiences conditioned on the current context instead of replaying them verbatim, improving performance on ALFWorld and WebShop while reducing negative transfer.
MemLens is a new benchmark for evaluating memory capabilities in large vision-language models through multi-session conversations. It compares long-context and memory-augmented approaches, revealing limitations in both and motivating hybrid architectures.