Interpretable Graph Capsule Networks for Object Recognition

Capsule Networks, as alternatives to Convolutional Neural have been proposed recognize objects from images. The current literature demonstrates many advantages of CapsNets over CNNs. However, how create explanations for individual classifications has not well explored. widely used saliency methods are mainly explaining CNN-based classifications; they map by combining activation values and the corresponding gradients, e.g., Grad-CAM. These require a specific architecture underlying classifiers cannot be trivially applied due iterative routing mechanism therein. To overcome lack interpretability, we can either propose new post-hoc interpretation or modifying model build-in explanations. In this work, explore latter. Specifically, interpretable Graph Networks (GraCapsNets), where replace part with multi-head attention-based Pooling approach. model, classification created effectively efficiently. Our also some unexpected benefits, even though it replaces fundamental CapsNets. GraCapsNets achieve better performance fewer parameters adversarial robustness, when compared Besides, keep other CapsNets, namely, disentangled representations affine transformation robustness.