FadeMem: Biologically-Inspired Forgetting for Efficient Agent Memory

Large language models deployed as autonomous agents face critical memory limitations, lacking selective forgetting mechanisms that lead to either catastrophic forgetting at context boundaries or information overload within them. While human memory naturally balances retention and forgetting through adaptive decay processes, current AI systems employ binary retention strategies that preserve everything or lose it entirely. We propose FadeMem, a biologically-inspired agent memory architecture that incorporates active forgetting mechanisms mirroring human cognitive efficiency. FadeMem implements differential decay rates across a dual-layer memory hierarchy, where retention is governed by adaptive exponential decay functions modulated by semantic relevance, access frequency, and temporal patterns. Through LLM-guided conflict resolution and intelligent memory fusion, our system consolidates related information while allowing irrelevant details to fade. Experiments on Multi-Session Chat, LoCoMo, and LTI-Bench demonstrate superior multi-hop reasoning and retrieval with 45% storage reduction, validating the effectiveness of biologically-inspired forgetting in agent memory systems.

💡 Research Summary

FadeMem addresses a critical bottleneck in large language model (LLM)‑driven autonomous agents: the lack of selective forgetting. Existing agent memory systems either retain all information indiscriminately until a hard context limit forces eviction, or they discard entire windows wholesale, leading to catastrophic forgetting. Inspired by human memory—where recall strength decays exponentially but is modulated by repetition, relevance, and emotional salience—the authors propose a biologically‑inspired architecture that actively forgets unimportant data while preserving and consolidating crucial knowledge.

The core of FadeMem is a dual‑layer hierarchy: a Short‑Term Memory Layer (SML) for low‑importance items and a Long‑Term Memory Layer (LML) for high‑importance items. Each memory entry mᵢ(t) is represented by a content embedding cᵢ, the original text sᵢ, a dynamic strength vᵢ(t)∈