PROJECT TITLE :

A Generative Model for Sparse Hyperparameter Determination - 2018

ABSTRACT:

Sparse autoencoder is an unsupervised feature extractor and has been widely used in the machine learning and knowledge mining community. However, a sparse hyperparameter needs to be determined to balance the trade-off between the reconstruction error and the sparsity of sparse autoencoder. Traditional sparse hyperparameter determination method is time-consuming, especially when the dataset is giant. In this Project, we tend to derive a generative model for sparse autoencoder. Based on this model, we tend to derive a formulation to determine the sparse hyperparameter effectively and efficiently. The relationship between the sparse hyperparameter and the common activation of sparse autoencoder hidden units is additionally presented in this Project. Experimental results and comparative studies over varied datasets demonstrate the effectiveness of our method to work out the sparse hyperparameter.