Learn Like Humans: Use Meta-cognitive Reflection for Efficient Self-Improvement

TL;DR

This paper introduces MARS, a framework inspired by human learning that enables large language models to self-improve efficiently in a single step by combining principle-based and procedural reflection, reducing computational costs.

Contribution

MARS is the first framework to achieve effective self-evolution in LLMs within a single recurrence cycle, inspired by educational psychology principles.

Findings

Outperforms state-of-the-art self-evolving systems on six benchmarks.

Reduces computational overhead compared to multi-turn recursive approaches.

Effectively refines reasoning logic without online feedback.

Abstract

While Large Language Models (LLMs) enable complex autonomous behavior, current agents remain constrained by static, human-designed prompts that limit adaptability. Existing self-improving frameworks attempt to bridge this gap but typically rely on inefficient, multi-turn recursive loops that incur high computational costs. To address this, we propose Metacognitive Agent Reflective Self-improvement (MARS), a framework that achieves efficient self-evolution within a single recurrence cycle. Inspired by educational psychology, MARS mimics human learning by integrating principle-based reflection (abstracting normative rules to avoid errors) and procedural reflection (deriving step-by-step strategies for success). By synthesizing these insights into optimized instructions, MARS allows agents to systematically refine their reasoning logic without continuous online feedback. Extensive experiments on six benchmarks demonstrate that MARS outperforms state-of-the-art self-evolving systems while significantly reducing computational overhead.