Jinyuan Zhang — Page 2

arXiv2023

Enabling surrogate-assisted evolutionary reinforcement learning via policy embedding

Evolutionary Reinforcement Learning (ERL) that applying Evolutionary Algorithms (EAs) to optimize the weight parameters of Deep Neural Network (DNN) based policies has been widely regarded as an alternative to traditional reinforcement learning methods. However, the evaluation of the iteratively generated population usually requires a large amount of computational time and can be prohibitively expensive, which may potentially restrict the applicability of ERL. Surrogate is often used to reduce the computational burden of evaluation in EAs. Unfortunately, in ERL, each individual of policy usually represents millions of weights parameters of DNN. This high-dimensional representation of policy has introduced a great challenge to the application of surrogates into ERL to speed up training. This paper proposes a PE-SAERL Framework to at the first time enable surrogate-assisted evolutionary reinforcement learning via policy embedding (PE). Empirical results on 5 Atari games show that the proposed method can perform more efficiently than the four state-of-the-art algorithms. The training process is accelerated up to 7x on tested games, comparing to its counterpart without the surrogate and PE.

cs.NE↗

arXiv2017

Preselection via Classification: A Case Study on Evolutionary Multiobjective Optimization

In evolutionary algorithms, a preselection operator aims to select the promising offspring solutions from a candidate offspring set. It is usually based on the estimated or real objective values of the candidate offspring solutions. In a sense, the preselection can be treated as a classification procedure, which classifies the candidate offspring solutions into promising ones and unpromising ones. Following this idea, we propose a classification based preselection (CPS) strategy for evolutionary multiobjective optimization. When applying classification based preselection, an evolutionary algorithm maintains two external populations (training data set) that consist of some selected good and bad solutions found so far; then it trains a classifier based on the training data set in each generation. Finally it uses the classifier to filter the unpromising candidate offspring solutions and choose a promising one from the generated candidate offspring set for each parent solution. In such cases, it is not necessary to estimate or evaluate the objective values of the candidate offspring solutions. The classification based preselection is applied to three state-of-the-art multiobjective evolutionary algorithms (MOEAs) and is empirically studied on two sets of test instances. The experimental results suggest that classification based preselection can successfully improve the performance of these MOEAs.

cs.NE↗