Fortunate Recall: Ontology-Driven Memory Lifecycle Management for Persistent Coherence in LLMs
Fortunate Recall argues that memory systems need lifecycle rules, not just larger stores.
The paper proposes a policy layer that classifies personal facts into behavioral categories, then applies decay, replacement, validity, and routing rules from extracted metadata. Its FR-Bank implementation beats several memory baselines on LifecycleBench and LongMemEval-S while cutting confabulation rates. The authors say ablations show generic lifecycle metadata drives correctness, while the ontology mainly improves calibration. They also report transfer to the BEAM benchmark and say the ontology, benchmark, and code are released. ArXiv · AI/CL/LG's note
The paper proposes a policy layer that classifies personal facts into behavioral categories, then applies decay, replacement, validity, and routing rules from extracted metadata. Its FR-Bank implementation beats several memory baselines on LifecycleBench and LongMemEval-S while cutting confabulation rates. The authors say ablations show generic lifecycle metadata drives correctness, while the ontology mainly improves calibration. They also report transfer to the BEAM benchmark and say the ontology, benchmark, and code are released. ArXiv · AI/CL/LG's note
score 5