D-MEM: Dopamine-Gated Agentic Memory via Reward Prediction Error Routing

Researchers introduce D-MEM, a biologically-inspired memory architecture for AI agents that uses dopamine-like reward prediction error routing to dramatically reduce computational costs. The system reduces token consumption by over 80% and eliminates quadratic scaling bottlenecks by selectively processing only high-importance information through cognitive restructuring.

• →D-MEM uses a Fast/Slow routing system based on Reward Prediction Error to decouple routine interactions from memory restructuring.
• →The architecture reduces token consumption by over 80% compared to existing append-and-evolve memory systems.
• →A lightweight Critic Router evaluates stimuli for Surprise and Utility to determine processing priority.
• →The system eliminates O(N²) write-latency bottlenecks that plague current autonomous LLM agent memory systems.
• →Performance evaluations show superior multi-hop reasoning and adversarial resilience using the new LoCoMo-Noise benchmark.