Memory Consolidation

Memory consolidation studies how learners turn transient experiences into durable knowledge while balancing adaptation, retention, storage, and access. Approaches include sleep-like replay and fast/slow synaptic dynamics in continual learning, and episodic-to-semantic transfer, graph or topic organization, and learned compression in language-agent memory. Work is evaluated in settings such as Split-MNIST and Split-CIFAR continual learning, and long-term recall benchmarks including LongMemEval and LoCoMo. Consolidation can save capacity and improve generalization, but may discard useful episodes, provenance, or temporal cues, leaving an open tradeoff between compact summaries and preserving original evidence.