An Improved Auto-encoder Based on 2-Level Prioritized Experience Replay for High Dimension Skewed Data

Abstract

Auto-encoder as the representative method for data dimensionality reduction and feature extraction, plays a very important role on machine learning. However, the data in the actual research work or industrial production are not always normalized data, at this time, it will lead high reconstruction error and slow convergence speed. This study proposed an improved auto-encoder and a denoising auto-encoder based on 2-level prioritized experience replay, which can improve accuracy and reduce loss, while processing a dimensionality reduction or feature extraction problem on high dimension skewed data. In order to evaluate the effectiveness of the proposed method, three models of high dimension simulation dataset which on different skewed degrees are generated. The results of evaluation experiments show that the proposed method can get lower reconstruction error than conventional method for high dimension skewed simulation data.

Xin Li

Lecturer

Research in areas such as machine learning, deep learning, and artificial intelligence applications, specialize in priority sampling and learning algorithm for non-uniform data and offset data.Learn more