SEDIMA: Cross-Run Hierarchical Insight Memory for Evolutionary Search Agents

2026-10-01

AI Summary

This study addresses the lack of memory and the tendency toward redundant exploration in LLM-based evolutionary search agents. To overcome these limitations, we propose a cross-run hierarchical insight memory mechanism. This approach constructs persistent memory through natural language distillation, attention-weighted clustering, and semantic retrieval, enabling agents to accumulate transferable knowledge across multiple runs that guides mutation without requiring modifications to existing operators. Experimental results demonstrate that the proposed mechanism yields average performance improvements of 5.5%–6.6% and reduces the number of iterations required to reach benchmarks by 32.3%.

Abstract

Large language model (LLM)-driven evolutionary search is a powerful paradigm for automated program and algorithm discovery, yet existing systems are largely memoryless: each run explores from scratch, so agents repeatedly rediscover the same improvements and re-encounter the same dead ends. We introduce SEDIMA, a persistent hierarchical insight memory for evolutionary search agents. SEDIMA distills raw traces into natural-language insights, clusters them by semantic similarity using attention-weighted centroids, and retrieves relevant guidance to condition future mutations, accumulating transferable knowledge across runs and problems rather than within a single trajectory. As a drop-in module that leaves the search operators unmodified, SEDIMA improves average final performance by 5.5% on AlgoTune and 6.6% on ALE-Bench LITE under a fixed budget of 100 evaluated candidates. Under OpenEvolve, SEDIMA requires 32.3% fewer iterations on average to reach baseline-best performance across the five evaluated backbones.

Problem

• evolutionary search
• large language models
• memoryless systems
• knowledge accumulation
• automated discovery

Innovation

• Evolutionary Search
• Hierarchical Insight Memory
• Large Language Models
• Cross-Run Knowledge Transfer
• Semantic Clustering