Gradient Search to Maximize Likelihood in ANN

• The cross entropy can be used in a gradient ascent algorithm for ANNs
• 
• , where is the th input to unit for the th training example
• The rule that minimizes squared error seeks under the assumptions that the training data can be modeled by Normally distributed noise added to a target function value.
• The rule that minimizes cross entropy sees under the assumption that the observed boolean value is a probabilistic function of the input instance.