Adaptive continuous action-set learning automata scheme

Authors：Guo Ying; Li Shenghong^*; Fan Bo

Source：Electronics Letters, 2018, 54(4): 242-U56.

DOI：10.1049/el.2017.4014

• learning automata
• sampling methods
• iterative methods
• adaptive continuous action-set learning automata scheme
• ACALA algorithm
• classic Beigy algorithm
• sampling process
• iterating process
• noisy function optimisation
• general evaluation system
• noise-added scenarios
• global optimum

Learning automata (LA) has been widely encountered in signal processing for images, speeches, videos and so on. One special type of LA, i.e. the CALA (continuous action-set LA) issue is handled. It first fills the existing gap aimed at the classic Beigy's algorithm, based on which a novel adaptive CALA (ACALA) algorithm is proposed. The proposed ACALA algorithm includes a sampling process and an iterating process, where the learning parameters are adaptively adjusted, not only between the two stages but also inside the core stage. Experiments with regard to noisy function optimisation reveal its outperformance, in accordance with the general evaluation system built. Specifically, the final result of the proposed algorithm could always judge as the global optimum even in noise-added scenarios, regardless of the initial parameters.

• Institution
  上海交通大学