Embedded Encoder-Decoder in Convolutional Networks Towards Explainable AI

Understanding intermediate layers of a deep learning model and discovering\nthe driving features of stimuli have attracted much interest, recently.\nExplainable artificial intelligence (XAI) provides a new way to open an AI\nblack box and makes a transparent and interpretable decision. This paper\nproposes a new explainable convolutional neural network (XCNN) which represents\nimportant and driving visual features of stimuli in an end-to-end model\narchitecture. This network employs encoder-decoder neural networks in a CNN\narchitecture to represent regions of interest in an image based on its\ncategory. The proposed model is trained without localization labels and\ngenerates a heat-map as part of the network architecture without extra\npost-processing steps. The experimental results on the CIFAR-10, Tiny ImageNet,\nand MNIST datasets showed the success of our algorithm (XCNN) to make CNNs\nexplainable. Based on visual assessment, the proposed model outperforms the\ncurrent algorithms in class-specific feature representation and interpretable\nheatmap generation while providing a simple and flexible network architecture.\nThe initial success of this approach warrants further study to enhance weakly\nsupervised localization and semantic segmentation in explainable frameworks.\n