Overfitting-based Bias-generating Neural-network : Simulation of Pavlov’s Dog

Since the advent of artificial intelligence techniques, many scholars have considered the phenomenon of overfitting to be negative and have worked to eliminate it. However, there are still many scholars who believe that the phenomenon of overfitting is worth exploiting. This paper argues that the phenomenon of overfitting of neural networks can reproduce the process of biologically generated prejudice, and builds a neural network prone to overfitting and applies it to a simple model of a dog. The simulations showed that the behavior of the simulated dog was similar to the conditioned reflexes in Pavlov’s dog experiments. Therefore, this paper argues that the judicious use of overfitting techniques can better enable artificial intelligence to be “each in its way”.