7 memory layers
Working, Short-Term, Episodic, Semantic, Procedural, Core, and Predictive Memory — each with its own time scale, persistence, and consolidation logic.
Track A
„How can AI remember, reason, and learn over time — without losing efficiency, reliability, or methodological traceability along the way?"
A seven-layer memory architecture grounded in neuroscience and tested in a productive application.
Motivation
Today's large language models forget. Their context windows are technically bounded, and plain vector stores are not enough to faithfully model the layers of human memory — from a fleeting working memory to consolidated long-term knowledge.
What is missing are architectures that explicitly model consolidation, directed forgetting, confidence propagation, and predictive memory formation. This research answers with an architecture that did not emerge from marketing but from the question how a system at an academic level can maintain its own knowledge base.
Methodological approach
The architecture splits memory into seven functionally differentiated layers — from working memory for active tasks, through episodic memory for concrete experiences, to predictive memory with neuromodulator-driven consolidation. Each layer is addressable through a unified API; a MemoryCoordinator orchestrates storing, retrieving, consolidating, and forgetting.
Fifteen algorithms — each with a peer-reviewed neuroscience source — provide the mechanics: FSRS for directed spaced repetition, Hebbian learning for co-activation, the Ebbinghaus forgetting curve, Bayesian confidence propagation, a two-factor synaptic model, a vmPFC-coupled prediction-error update, and six more. Sleep consolidation simulates offline memory replay.
Core components
Working, Short-Term, Episodic, Semantic, Procedural, Core, and Predictive Memory — each with its own time scale, persistence, and consolidation logic.
A unified API across all seven layers: store(), recall(), consolidate(), decay(), reviewQueue(). Routing follows salience and context automatically.
Free Spaced Repetition Scheduler — vmPFC-coupled variant with prediction-error update. Facts are reviewed at optimal moments, not on a fixed grid.
Offline memory replay as a methodological core. Selection of important memories for reinforcement; directed pruning of weak associations.
Consolidation is driven by two factors — co-activation frequency and a neuromodulator signal. Reproduces experimental findings from the synaptic-tagging literature.
Six components — NeuromodulatorEngine, ReconsolidationEngine, TripleCopyMemory, PriorityMap, StabilityProtector, MetacognitiveMonitor — model predictive memory formation with bias detection.
Get involved
Enquiries from universities, research consortia or funding programmes — on replication studies, consortium proposals or methodological discussion — are explicitly welcome.
research@zensation.aiMore from this research
Three tracks, one core, a shared ethics foundation.