Visualizing the Emergence of Intermediate Visual Patterns in DNNs
This provides insights into DNN behavior for researchers, with applications to techniques like adversarial attacks and knowledge distillation, but it is incremental as it builds on existing visualization methods.
The paper tackles the problem of understanding how deep neural networks (DNNs) learn visual patterns by proposing a method to visualize the emergence of discriminative patterns in intermediate layers during training, and it quantifies knowledge points to evaluate representation capacity.
This paper proposes a method to visualize the discrimination power of intermediate-layer visual patterns encoded by a DNN. Specifically, we visualize (1) how the DNN gradually learns regional visual patterns in each intermediate layer during the training process, and (2) the effects of the DNN using non-discriminative patterns in low layers to construct disciminative patterns in middle/high layers through the forward propagation. Based on our visualization method, we can quantify knowledge points (i.e., the number of discriminative visual patterns) learned by the DNN to evaluate the representation capacity of the DNN. Furthermore, this method also provides new insights into signal-processing behaviors of existing deep-learning techniques, such as adversarial attacks and knowledge distillation.