A Cognitive State Identification Method Based on Dynamic Q-Leaming Model Parameters

Exploring the changes of animal cognitive states in behavioral decision-making tasks can not only be used to evaluate its learning ability under certain conditions, but also has important significance for the ethology research. In this study, pigeons were trained to perform the Markov decision task with two time-steps and fixed state transition probability, Q-learning model was used to characterize the process of cognitive state change in experimental animals, and a dynamic identification method of key parameters of Q-learning model based on PSO algorithm was proposed. The results shows that our method can effectively and finely identify the cognitive state of pigeons in behavioral decision-making tasks by characterizing the ability to weigh past experience and current experience and the consideration of future states, and it also can be used as a quantitative method to evaluate the learning ability of animals in specific tasks.