Visual analysis system for convolutional neural network based classifiers

摘要

A visual analysis method and system is disclosed for visualizing an operation of an image classification model having at least one convolutional neural network layer. The image classification model classifies sample images into one of a given set of possible classes. The visual analysis method determines a unified sorting of the given set of possible classes based on a similarity hierarchy so that classes that are similar to each other are classified together in the unified sort. The visual analysis method displays various graphical representations, including a class hierarchy viewer, a confusion matrix, and a reaction map. In any case, the elements of the unified sorting plots are arranged accordingly. Using the method, a user can better understand the training process of the model, diagnose the separation performance of the various feature detectors of the model, and enhance the architecture of the model.