Search papers, labs, and topics across Lattice.
This paper introduces ScrubJay-MEM, a memory architecture for LLM agents that operationalizes type-conditioned temporal decay based on principles from scrub jay episodic memory. By encoding each memory as a What-Where-When tuple with an associated perishability coefficient and utility horizon, the system effectively retrieves and updates memories while minimizing contamination from outdated facts. The Temporal Generalization Test reveals that ScrubJay-MEM significantly outperforms existing retrieval-based systems, achieving a positive Generalization Gap and notable improvements in performance metrics on various benchmarks.
Type-conditioned memory decay can enhance LLM performance by ensuring that only the most relevant and timely information is retrieved, leading to a significant boost in temporal reasoning capabilities.
LLM agents that persist across sessions accumulate stored memories whose validity varies enormously by content type, yet existing memory architectures treat all memories as equally persistent and systematically contaminate retrieved context with outdated facts. We show that per-memory, type-conditioned temporal decay, a property of western scrub jay episodic memory, can be operationalized as an auto-classified coefficient $\pi_i$ in an external LLM-agent memory store, yielding ScrubJay-MEM: each memory is encoded as a jointly-bound What--Where--When tuple with an estimated perishability $\pi_i$ and utility horizon $\tau_i$, retrieved by query-adaptive scoring, and revised retroactively at $O(1)$ LLM calls per update. We introduce the Temporal Generalization Test (TGT), a benchmark with held-out retention intervals and a Generalization Gap (GenGap) metric. On TGT, ScrubJay-MEM is the only retrieval-based system with substantially positive GenGap ($+0.108$); on MemoryAgentBench EventQA-64k it improves F1 by $+2.66$ over Mem0 and $+3.09$ over Qwen3-Embedding-4B under a llm backbone. A decay ablation collapses GenGap by $5.7\times$, establishing type-conditioned decay as necessary for the result. Gains narrow under stronger backbones and reverse on fact-consolidation tasks, scoping the contribution to temporal reasoning over perishable facts.
