Visualizing the decision-making process in deep neural decision forest
This work provides interpretability tools for deep neural decision forests, which is an incremental improvement for researchers and practitioners in computer vision.
The authors tackled the problem of interpreting deep neural decision forests by visualizing their decision-making process and saliency maps for classification and regression tasks, and demonstrated the distribution of routing probabilities in a multi-task coordinate regression problem.
Deep neural decision forest (NDF) achieved remarkable performance on various vision tasks via combining decision tree and deep representation learning. In this work, we first trace the decision-making process of this model and visualize saliency maps to understand which portion of the input influence it more for both classification and regression problems. We then apply NDF on a multi-task coordinate regression problem and demonstrate the distribution of routing probabilities, which is vital for interpreting NDF yet not shown for regression problems. The pre-trained model and code for visualization will be available at https://github.com/Nicholasli1995/VisualizingNDF