Discrete space reinforcement learning algorithm based on twin support vector machine classification

Reinforcement learning (RL) has become one of the key component of various machine learning algorithms in recent years. However, traditional RL algorithms lack convergence speed and accuracy in small-scale discrete space environment. Recently An et al. proposed RL algorithm based on support vector machines (SVMs) (Pattern Recognit. Lett. 111 (2018) 30-35) which adopts the Advantage Actor-Critic (A2C) framework and improves the speed and accuracy of convergence in discrete space. Owing to the advantages of twin support vector machines (TWSVMs) over SVMs, in this paper, we propose a RL algorithm based on TWSVM classification. The proposed algorithm adopts a modified A2C framework, where there are multiple Actors and a single Critic. Finally, we compare the performance of the proposed algorithm with some existing algorithms in traditional RL environment. Interestingly, the proposed algorithm outperforms the existing algorithms in terms of convergence speed and accuracy. (c) 2022 Elsevier B.V. All rights reserved.