A new tool called CNN Explainer allows non-experts to learn how convolutional neural networks (CNNs) work and examine their structure in an interactive manner.
Group of researchers has conducted a number of interviews and surveys with deep learning instructors and past students to identify the key challenges that novices face when learning about convolutional neural networks. The results and outcomes from these surveys were taken into account when building CNN Explainer, whose goal is to help learners overcome difficulties.
The tool integrates an overview of a convolutional neural network with dynamic explanation views of different components of a network, and all that in an interactive web-based implementation. The demo of CNN Explainer available here contains several example images that can be interactively fed through a few-layer CNN network for classification. The different feature maps at each layer are shown as well as interactive views where kernels or filters are passed over an input feature map.
Researchers mention that CNN Explainer as a learning tool can help many beginners in the field to better understand the underlying mechanisms of convolutional neural networks and speed up their learning. The tool runs in a web-browser without the need for any installation making it accessible to a wide range of people that want to learn about CNNs.
The results from the surveys together with the design of CNN Explainer can be found in the paper published on arxiv. The implementation of the tool was open-sourced and is available on Github. The demo can be accessed here.