EmoDM: A Diffusion Model for Evolutionary Multi-objective Optimization

Xueming Yan,Yaochu Jin

2024-08-15

Abstract:Evolutionary algorithms have been successful in solving multi-objective optimization problems (MOPs). However, as a class of population-based search methodology, evolutionary algorithms require a large number of evaluations of the objective functions, preventing them from being applied to a wide range of expensive MOPs. To tackle the above challenge, this work proposes for the first time a diffusion model that can learn to perform evolutionary multi-objective search, called EmoDM. This is achieved by treating the reversed convergence process of evolutionary search as the forward diffusion and learn the noise distributions from previously solved evolutionary optimization tasks. The pre-trained EmoDM can then generate a set of non-dominated solutions for a new MOP by means of its reverse diffusion without further evolutionary search, thereby significantly reducing the required function evaluations. To enhance the scalability of EmoDM, a mutual entropy-based attention mechanism is introduced to capture the decision variables that are most important for the objectives. Experimental results demonstrate the competitiveness of EmoDM in terms of both the search performance and computational efficiency compared with state-of-the-art evolutionary algorithms in solving MOPs having up to 5000 decision variables. The pre-trained EmoDM is shown to generalize well to unseen problems, revealing its strong potential as a general and efficient MOP solver.

Neural and Evolutionary Computing,Artificial Intelligence

What problem does this paper attempt to address?

The paper aims to address a key challenge encountered in multi-objective optimization problems (MOPs): the fact that most evolutionary algorithms (MOEAs) require a large number of objective function evaluations, which limits their application in solving real-world problems with high computational costs.

To tackle the aforementioned challenge, the authors propose a new model called EmoDM (Evolutionary Multi-objective Diffusion Model). EmoDM is a diffusion model-based approach, whose core idea is to learn the noise distribution from previously solved multi-objective optimization tasks and treat the reverse process of evolutionary search as forward diffusion. Specifically, EmoDM achieves its goal through the following steps:

1. **Model Training**: EmoDM is trained using data obtained from multiple solved multi-objective optimization problems, which includes the changes in solutions between different generations. During training, the model learns to estimate the noise distribution from forward diffusion (i.e., the reverse evolutionary search process).

2. **Solution Generation**: Once training is complete, EmoDM can generate an approximate Pareto optimal set starting from a randomly generated initial population through the reverse diffusion process, without the need for additional evolutionary search processes, thereby significantly reducing the number of function evaluations required.

To enhance EmoDM's ability to handle high-dimensional problems, the authors also introduce a mutual entropy-based attention mechanism to capture the most important parts of the decision variables for the objectives, enabling the model to focus on the most relevant features.

Experimental results show that compared to four state-of-the-art evolutionary algorithms, EmoDM not only performs excellently in search performance but also demonstrates competitive computational efficiency, especially when solving complex multi-objective optimization problems with up to 5000 decision variables. Additionally, the pre-trained EmoDM model can generalize well to unseen problems, showcasing its strong potential as an efficient and general-purpose multi-objective optimization solver.