Achieves a +48pp accuracy gain in agents using a non-parametric online learning framework that reuses procedural plans without updating model weights.
April 1, 2026
Original Paper
APEX-EM: Non-Parametric Online Learning for Autonomous Agents via Structured Procedural-Episodic Experience Replay
arXiv · 2603.29093
The Takeaway
It allows LLM agents to build a 'procedural memory' of how they solved previous tasks, enabling them to handle structurally similar but lexically different problems. This solves the 'start from scratch' problem in autonomous agents without the instability of fine-tuning.
From the abstract
LLM-based autonomous agents lack persistent procedural memory: they re-derive solutions from scratch even when structurally identical tasks have been solved before. We present \textbf{APEX-EM}, a non-parametric online learning framework that accumulates, retrieves, and reuses structured procedural plans without modifying model weights. APEX-EM introduces: (1) a \emph{structured experience representation} encoding the full procedural-episodic trace of each execution -- planning steps, artifacts,