Large Language Model-Powered Evolutionary Code Optimization on a Phylogenetic Tree

2026-01-20

AI Summary

This work proposes PhyloEvolve, a context-aware reinforcement learning–based LLM agent system designed to address the labor-intensive and inefficient iterative nature of manual optimization in modern GPU-accelerated scientific computing. The approach formulates algorithmic optimization as a sequential decision-making problem, leveraging code modifications and performance feedback as learning signals. It innovatively organizes the optimization history using a phylogenetic tree structure, enabling backtracking, cross-lineage knowledge transfer, and reproducibility. By integrating algorithm distillation with a prompt-driven Decision Transformer, PhyloEvolve achieves trajectory-conditioned experience reuse without requiring retraining. Evaluated on tasks including PDE solvers, manifold learning, and spectral graph algorithms, PhyloEvolve demonstrates significant improvements over baseline and conventional evolutionary methods in runtime efficiency, memory usage, and correctness.

Technology Category

• Search and Optimization: Evolutionary Computation
• Machine Learning: Evolutionary Learning
• Multiagent Systems: Multiagent Learning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Optimizing scientific computing algorithms for modern GPUs is a labor-intensive and iterative process involving repeated code modification, benchmarking, and tuning across complex hardware and software stacks. Recent work has explored large language model (LLM)-assisted evolutionary methods for automated code optimization, but these approaches primarily rely on outcome-based selection and random mutation, underutilizing the rich trajectory information generated during iterative optimization. We propose PhyloEvolve, an LLM-agent system that reframes GPU-oriented algorithm optimization as an In-Context Reinforcement Learning (ICRL) problem. This formulation enables trajectory-conditioned reuse of optimization experience without model retraining. PhyloEvolve integrates Algorithm Distillation and prompt-based Decision Transformers into an iterative workflow, treating sequences of algorithm modifications and performance feedback as first-class learning signals. To organize optimization history, we introduce a phylogenetic tree representation that captures inheritance, divergence, and recombination among algorithm variants, enabling backtracking, cross-lineage transfer, and reproducibility. The system combines elite trajectory pooling, multi-island parallel exploration, and containerized execution to balance exploration and exploitation across heterogeneous hardware. We evaluate PhyloEvolve on scientific computing workloads including PDE solvers, manifold learning, and spectral graph algorithms, demonstrating consistent improvements in runtime, memory efficiency, and correctness over baseline and evolutionary methods. Code is published at: https://github.com/annihi1ation/phylo_evolve

Problem

• GPU code optimization
• scientific computing
• evolutionary optimization
• trajectory reuse
• large language models

Innovation

• PhyloEvolve
• In-Context Reinforcement Learning
• Algorithm Distillation
• Phylogenetic Tree
• LLM-Agent Optimization