Towards interpreting computer vision based on transformation invariant optimization
This work addresses the challenge of understanding DNN predictions for researchers and practitioners, but it is incremental as it builds on existing back-propagation methods.
The authors tackled the problem of interpreting deep neural networks (DNNs) in computer vision by generating visualized images that activate target classes using back-propagation, with rotation and scaling operations improving visualization effect.
Interpreting how does deep neural networks (DNNs) make predictions is a vital field in artificial intelligence, which hinders wide applications of DNNs. Visualization of learned representations helps we humans understand the vision of DNNs. In this work, visualized images that can activate the neural network to the target classes are generated by back-propagation method. Here, rotation and scaling operations are applied to introduce the transformation invariance in the image generating process, which we find a significant improvement on visualization effect. Finally, we show some cases that such method can help us to gain insight into neural networks.