Ye Zhao, phd

Ye Zhao is a professor in the Department of Computer Science at Kent State University, Ohio, USA. He has been working on computer graphics and visualization for more than 20 years. His current research interests include visual analytics of urban data, multidimensional, text, and multimedia data visualization. He has published numerous refereed technical papers and served in many program committees of data visualization conferences. Ye Zhao received his PhD degree in computer science from Stony Brook University in 2006.

Abstract

Visual Design of Layer-wise Relevance Propagation Models for Deep Learning Explanation

Layer-wise Relevance Propagation (LRP) methods are widely used for interpreting the prediction results of convolutional neural networks (CNN). Multiple LRP variations utilize a set of relevance backpropagation rules with various parameters. Moreover, composite LRPs apply different rules on segments of CNN layers. These features impose great challenge for users to design, explore, and find suitable LRP models. We develop a visual model designer, named as VisLRP, which helps LRP designers and students efficiently perform these tasks. Various LRPs are unified into an integrated framework with an intuitive workflow of parameter setup. Therefore, VisLRP allows users to interactively configure LRP models, change parameters, and then study the relevance information. Moreover, VisLRP facilitates relevance based visual analysis with two important functions: relevance-based pixel flipping and neuron ablation.