Learning Automata
Junqi Zhang, MengChu Zhou · 2023
Learning Automaton (LA) is one of the reinforcement learning algorithms and has the advantages of being simple and easy to implement, possessing fast random optimization and strong anti-noise ability, and complete convergence. In addition, LAs have also been applied to many application fields, such as graph coloring, random shortest path, wireless network spectrum allocation, image processing, and pattern recognition. An LA consists of two parts: an automaton and an environment. There are several common fixed-structure automata, i.e., Tsetlin, Krylov, Krinsky, Iraji–Jamalian Automaton (IJA), and Tunable Fixed Structure Learning Automaton (TFSLA). In this chapter, the authors have introduced the fixed structure stochastic automata in stationary random environments. Their state transition probabilities and action probabilities are fixed. In terms of whether there is an estimator and whether the estimator is stochastic, LAs with a variable structure can be divided into estimator-free, deterministic estimator, and stochastic estimator LAs.