Evolutionary prompt optimization has demonstrated effectiveness in refining prompts for LLMs. However, existing approaches lack robust operators and efficient evaluation mechanisms. In this work, we propose several key improvements to evolutionary prompt optimization that can partially generalize to prompt optimization in general: 1) decomposing evolution into distinct steps to enhance the evolution and its control, 2) introducing an LLM-based judge to verify the evolutions, 3) integrating human feedback to refine the evolutionary operator, and 4) developing more efficient evaluation strategies that maintain performance while reducing computational overhead. Our approach improves both optimization quality and efficiency. We release our code, enabling prompt optimization on new tasks and facilitating further research in this area.

Evolutionary computation (EC), as a powerful optimization algorithm, has been applied across various domains. However, as the complexity of problems increases, the limitations of EC have become more apparent. The advent of large language models (LLMs) has not only transformed natural language processing but also extended their capabilities to diverse fields. By harnessing LLMs' vast knowledge and adaptive capabilities, we provide a forward-looking overview of potential improvements LLMs can bring to EC, focusing on the algorithms themselves, population design, and additional enhancements. This presents a promising direction for future research at the intersection of LLMs and EC.

This paper formulates model selection as an infinite-armed bandit problem. The models are arms, and picking an arm corresponds to a partial training of the model (resource allocation). The reward is the accuracy of the selected model after its partial training. In this best arm identification problem, regret is the gap between the expected accuracy of the optimal model and that of the model finally chosen. We first consider a straightforward generalization of UCB-E to the stochastic infinite-armed bandit problem and show that, under basic assumptions, the expected regret order is $T^{-\alpha}$ for some $\alpha \in (0,1/5)$ and $T$ the number of resources to allocate. From this vanilla algorithm, we introduce the algorithm Mutant-UCB that incorporates operators from evolutionary algorithms. Tests carried out on three open source image classification data sets attest to the relevance of this novel combining approach, which outperforms the state-of-the-art for a fixed budget.

Large Language Models (LLMs) excel in various tasks, but they rely on carefully crafted prompts that often demand substantial human effort. To automate this process, in this paper, we propose a novel framework for discrete prompt optimization, called EvoPrompt, which borrows the idea of evolutionary algorithms (EAs) as they exhibit good performance and fast convergence. To enable EAs to work on discrete prompts, which are natural language expressions that need to be coherent and human-readable, we connect LLMs with EAs. This approach allows us to simultaneously leverage the powerful language processing capabilities of LLMs and the efficient optimization performance of EAs. Specifically, abstaining from any gradients or parameters, EvoPrompt starts from a population of prompts and iteratively generates new prompts with LLMs based on the evolutionary operators, improving the population based on the development set. We optimize prompts for both closed- and open-source LLMs including GPT-3.5 and Alpaca, on 9 datasets spanning language understanding and generation tasks. EvoPrompt significantly outperforms human-engineered prompts and existing methods for automatic prompt generation by up to 25% and 14% respectively. Furthermore, EvoPrompt demonstrates that connecting LLMs with EAs creates synergies, which could inspire further research on the combination of LLMs and conventional algorithms.

We present a simple, sample-efficient algorithm for introducing large but directed learning steps in reinforcement learning (RL), through the use of evolutionary operators. The methodology uses a population of RL agents training with a common experience buffer, with occasional crossovers and mutations of the agents in order to search efficiently through the policy space. Unlike prior literature on combining evolutionary search (ES) with RL, this work does not generate a distribution of agents from a common mean and covariance matrix. Neither does it require the evaluation of the entire population of policies at every time step. Instead, we focus on gradient-based training throughout the life of every policy (individual), with a sparse amount of evolutionary exploration. The resulting algorithm is shown to be robust to hyperparameter variations. As a surprising corollary, we show that simply initialising and training multiple RL agents with a common memory (with no further evolutionary updates) outperforms several standard RL baselines.