Approximation of Ideal observer using GAN and MCMC
Ideal Observer (IO) is considered the ultimate observer comparing to all others when it comes to signal detection performance since it is based on the complete statistics of the background and signal. However, the computation of IO is generally considered intractable. Markov-Chain Monte Carlo method was used to simulate the distribution about two decades ago using simple models. GAN is also considered as the generative model that can be used to characterize the distribution of a stochastic background in a certain sense. Here we verify the performance of GAN about the characterization of distribution using the signal detection task and MCMC technique as a comparison.