Top K Learning Automata Algorithms

Machine Learning is divided into three categories: Supervised Learning,Unsupervised Learning and Reinforcement Learning.Learning Automata is one of the most important algorithms among Reinforcement Learning.By interacting with the environment,Learning Automata can find the optimal action without a priori knowledge of the environment.Learning Automata,due to their simple,complete theory and anti-jamming capability,are widely used in pattern recognition,communications and other fields.As the demand for practical applications,research on new Top K Learning Automata gradually began.Top K Learning Automata's goal is to find the optimal subset of actions from the action set of the environment,which is more suitable for some industrial environments and global optimization and other practical applications.In this context,this paper conducts research on Top K Learning Automata algorithms.Currently,the research on this area is still in the initial stage.Existing algorithms only extended single ordinary Learning Automata to find the optimal subset of actions.To enable the existing large number of ordinary Learning Automata to be applied to this new target,the paper analyzes the general Learning Automata algorithms and the mechanism of new Top K Learning Automata,thus proposes two extension framework to extend mainstream ordinary Learning Automata to Top K Learning Automata--Naive Top K extension framework and Integral Top K extension framework and gives the specific processes and examples of these two frameworks.Through simulation experiments,the paper presents the advantages and disadvantages of two extension frameworks.Secondly,this paper studies the differences between Top K Learning Automata and ordinary Learning Automata and proposes a new Top K Learning Automata algorithm--LELAK algorithm.The update process of this algorithm is theoretically more consistent with the purpose of Top K Learning Automata.Simulation results show that,LELAK converge faster than the existing Top K Learning Automata.Furthermore,since the number of iterations of Top K Learning Automata is relatively large,in order to further improve the efficiency,the paper implements parallel structure in LELAK algorithm.Through parallelling multiple Learning Automata,accelerating the interaction with the environment,the number of iterations is reduced.Simulation results show that,after being parallelled,LELAK's convergence speed is significantly improved.