Disclosed are a method, a computer readable medium, and a system for training a neural network. The method includes the steps of selecting an input pattern from a set of training data that includes input patterns and noisy target patterns, the input patterns and the noisy target patterns each corresponding to a latent, defect-free target pattern. The input pattern is processed by a neural network model to produce an output, and a noisy target pattern is selected from the set of training data, the noisy target patterns having a distribution relative to the latent, defect-free target pattern. The method further includes adjusting parameter values of the neural network model to reduce differences between the output and the noisy target pattern.
展开▼