AdaEvolve: Adaptive LLM Driven Zeroth-Order Optimization

2026-02-23

AI Summary

This work addresses the limitations of existing large language model (LLM)-based evolutionary search methods, which rely on static scheduling strategies and struggle to adapt to non-stationary dynamics during optimization, often leading to inefficient resource allocation and insufficient exploration of promising regions. To overcome these challenges, the authors formulate LLM-driven evolutionary optimization as a hierarchical adaptive optimization problem and introduce a three-tier cooperative mechanism: local dynamic modulation of exploration intensity, global resource allocation via a multi-armed bandit framework, and a meta-guidance module that automatically generates novel problem-solving strategies upon detection of search stagnation. Integrating LLMs as semantic mutation operators with zeroth-order optimization and adaptive control theory, the proposed approach significantly outperforms current open-source baselines across 185 diverse open-ended problems spanning combinatorial optimization, system optimization, and algorithm design.

Technology Category

• Search and Optimization: Evolutionary Computation
• Machine Learning: Evolutionary Learning
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

The paradigm of automated program generation is shifting from one-shot generation to inference-time search, where Large Language Models (LLMs) function as semantic mutation operators within evolutionary loops. While effective, these systems are currently governed by static schedules that fail to account for the non-stationary dynamics of the search process. This rigidity results in substantial computational waste, as resources are indiscriminately allocated to stagnating populations while promising frontiers remain under-exploited. We introduce AdaEvolve, a framework that reformulates LLM-driven evolution as a hierarchical adaptive optimization problem. AdaEvolve uses an "accumulated improvement signal" to unify decisions across three levels: Local Adaptation, which dynamically modulates the exploration intensity within a population of solution candidates; Global Adaptation, which routes the global resource budget via bandit-based scheduling across different solution candidate populations; and Meta-Guidance which generates novel solution tactics based on the previously generated solutions and their corresponding improvements when the progress stalls. We demonstrate that AdaEvolve consistently outperforms the open-sourced baselines across 185 different open-ended optimization problems including combinatorial, systems optimization and algorithm design problems.

Problem

• LLM-driven evolution
• zeroth-order optimization
• adaptive scheduling
• non-stationary dynamics
• computational waste

Innovation

• adaptive optimization
• LLM-driven evolution
• zeroth-order optimization
• hierarchical adaptation
• bandit-based scheduling