A neural network-based digital hearing aid intelligent fitting method and system

Abstract

The application provides a digital hearing aid intelligent fitting method and system of a neural network, and the method comprises the following steps: collecting fitting big data, inputting the fitting big data into a deep neural network model to obtain a basic fitting model through training; obtaining the weight parameters, learning rate and weight decay coefficient of the basic fitting model as a reinforcement learning strategy; inputting a reinforcement learning strategy of a random probability distribution into the current basic fitting model to output a corresponding reinforcement learning prediction value, and calculating the error between the prediction value and an actual value; configuring the reinforcement learning reward and punishment rules, calculating the feature eligibility and eligibility trace of the current reinforcement learning, calculating the corresponding update strategy according to the error between the prediction value and the actual value, the reward and punishment rules, the feature eligibility and the eligibility trace, and executing the next reinforcement learning action according to the update strategy until the current strategy meets the constraint condition.

Images