Unsupervised Discovery of Nonlinear Structure Using Contrastive Backpropagation

摘要: We describe a way of modeling high-dimensional data vectors by using an unsupervised, nonlinear, multilayer neural network in which the activity each neuron-like unit makes additive contribution to global energy score that indicates how surprised is vector. The connection weights determine depends on activities earlier layers are learned minimizing assigned actually observed and maximizing “confabulations” generated perturbing vector direction decreases its under current model.