Visualization hidden layer projection of Artificial Neural Networks

Short description

One of the research work includes extracting value of travel times from Artificial Neural Networks (ANN). We realized that these some of these extracted value of travel times from ANN were not making sense so I explored the cause. It is difficult to communicate how the hidden layer transformation are causing erroneous value of travel time. The explanations are beyond the scope of this article.I have shared one of the simpler plot and a simple video showing the process involved in VOTT estimation. These videos were used to understand the issues and to communicate with others.

Figure

Figure shows the contour plots of partial derivative used to estimate VOTT. The color of these contours explains the cause and were used to communicate with others

Animated video

The animated video was created to understand the estimation process of value of travel time in an Artificial Neural Network. It contains several subplots including partial derivatives and the transformed that which gave us very detailed insights of the process and problems involved in estimation.