Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An investigation on the use of Large Language Models for hyperparameter tuning in Evolutionary Algorithms

Leonardo Lucio Custode, Fabio Caraffini, Anil Yaman, Giovanni Iacca

TL;DR

The paper addresses the challenge of hyperparameter tuning in Evolutionary Computation by exploring automated step-size adaptation for the $(1+1)$-ES using two open-source LLMs, Llama2-70b and Mixtral. By querying the LLMs with optimization logs every $p=50$ generations and constraining the new step-size to $s \in [0.001,0.999]$, the authors evaluate whether LLMs can provide real-time, effective recommendations within a fixed budget of $10^3$ fitness evaluations on the BBOB/IOHProfiler suite. Results show that LLM-based strategies can be competitive with the traditional One-Fifth rule, with Llama2-70b delivering the most consistent performance across problem dimensions $2$, $5$, and $30$; Mixtral performs well at low dimensions but struggles as dimensionality increases. The work demonstrates the feasibility of LLM-guided hyperparameter tuning in Evolutionary Algorithms and suggests avenues for specialized, domain-tuned LLMs and larger-scale evaluations.

Abstract

Hyperparameter optimization is a crucial problem in Evolutionary Computation. In fact, the values of the hyperparameters directly impact the trajectory taken by the optimization process, and their choice requires extensive reasoning by human operators. Although a variety of self-adaptive Evolutionary Algorithms have been proposed in the literature, no definitive solution has been found. In this work, we perform a preliminary investigation to automate the reasoning process that leads to the choice of hyperparameter values. We employ two open-source Large Language Models (LLMs), namely Llama2-70b and Mixtral, to analyze the optimization logs online and provide novel real-time hyperparameter recommendations. We study our approach in the context of step-size adaptation for (1+1)-ES. The results suggest that LLMs can be an effective method for optimizing hyperparameters in Evolution Strategies, encouraging further research in this direction.

An investigation on the use of Large Language Models for hyperparameter tuning in Evolutionary Algorithms

TL;DR

The paper addresses the challenge of hyperparameter tuning in Evolutionary Computation by exploring automated step-size adaptation for the -ES using two open-source LLMs, Llama2-70b and Mixtral. By querying the LLMs with optimization logs every generations and constraining the new step-size to , the authors evaluate whether LLMs can provide real-time, effective recommendations within a fixed budget of fitness evaluations on the BBOB/IOHProfiler suite. Results show that LLM-based strategies can be competitive with the traditional One-Fifth rule, with Llama2-70b delivering the most consistent performance across problem dimensions , , and ; Mixtral performs well at low dimensions but struggles as dimensionality increases. The work demonstrates the feasibility of LLM-guided hyperparameter tuning in Evolutionary Algorithms and suggests avenues for specialized, domain-tuned LLMs and larger-scale evaluations.

Abstract

Hyperparameter optimization is a crucial problem in Evolutionary Computation. In fact, the values of the hyperparameters directly impact the trajectory taken by the optimization process, and their choice requires extensive reasoning by human operators. Although a variety of self-adaptive Evolutionary Algorithms have been proposed in the literature, no definitive solution has been found. In this work, we perform a preliminary investigation to automate the reasoning process that leads to the choice of hyperparameter values. We employ two open-source Large Language Models (LLMs), namely Llama2-70b and Mixtral, to analyze the optimization logs online and provide novel real-time hyperparameter recommendations. We study our approach in the context of step-size adaptation for (1+1)-ES. The results suggest that LLMs can be an effective method for optimizing hyperparameters in Evolution Strategies, encouraging further research in this direction.
Paper Structure (11 sections, 1 equation, 3 figures, 3 tables)

This paper contains 11 sections, 1 equation, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. LLMs as the optimizer
  4. EAs for prompt or LLM optimization
  5. LLMs supporting EAs
  6. Methods
  7. Benchmark problems
  8. Prompt format
  9. Algorithms and LLMs
  10. Results
  11. Conclusions

Figures (3)

  • Figure 1: A block diagram showing the self-adaptive step-size adaptation scheme studied in this work.
  • Figure 2: Average step-size for each function evaluation for all the methods under comparison. The solid lines represent the mean value for the step-size (averaged over all problems and problem dimensionalities), while the shaded area represents the 95% confidence interval.
  • Figure 3: Boxplot of the best fitness obtained in 10 runs by each of the methods on each function from the BBOB suite. Note that we aggregate the values from $10$ runs on all the considered problem sizes.