A walk in the black-box: 3D visualization of large neural networks in virtual reality

Abstract

Within the last decade Deep Learning has become a tool for solving challenging problems like image recognition. Still, Convolutional Neural Networks (CNNs) are considered black-boxes, which are difficult to understand by humans. Hence, there is an urge to visualize CNN architectures, their internal processes and what they actually learn. Previously, virtual realityhas been successfully applied to display small CNNs in immersive 3D environments. In this work, we address the problem how to feasibly render large-scale CNNs, thereby enabling the visualization of popular architectures with ten thousands of feature maps and branches in the computational graph in 3D. Our software ”DeepVisionVR” enables the user to freely walk through the layered network, pick up and place images, move/scale layers for better readability, perform feature visualization and export the results. We also provide a novel Pytorch module to dynamically link PyTorch with Unity, which gives developers and researchers a convenient interface to visualize their own architectures. The visualization is directly created from the PyTorch class that defines the Pytorch model used for training and testing. This approach allows full access to the network’s internals and direct control over what exactly is visualized. In a use-case study, we apply the module to analyze models with different generalization abilities in order to understand how networks memorize images. We train two recent architectures, CovidResNet and CovidDenseNet on the Caltech101 and the SARS-CoV-2 datasets and find that bad generalization is driven by high-frequency features and the susceptibility to specific pixel arrangements, leading to implications for the practical application of CNNs. The code is available on Github https://github.com/Criscraft/DeepVisionVR.

Using Augmented Reality Interfaces for Artificial Intelligence Systems

“…Linse, Alshazly and Martinetz [11] applied visualization in virtual reality by addressing the problem of how to create complex CNNs.…”

Visualization of AI Systems in Virtual Reality: A Comprehensive Review

“…They used TensorFlow as their AI system to provide detailed information for neural networks and permit the adjustment of training parameters. Similarly, [12] employed the Unity game engine with OpenXR support for immersive visualization and interaction with convolutional neural networks (CNNs). Their AI system of choice was PyTorch, a deep learning framework designed to interconnect with Unity for dynamic visualization and interaction with custom CNN architectures.…”

“…The study [11] utilized this approach, enabling users to interact with each neuron in the network. Similarly, [12] used 3D rendering of Convolutional Neural Networks (CNNs), representing the computational graph as a connected conveyor system and optimizing the rendering of large architectures. They also allowed users to move, scale, and interact with CNN layers, offering a display of weight distributions, classification results, and feature visualizations.…”

“…The immersive experience of VR can offer new possibilities to make Convolutional Neural Networks (CNNs) more human-understandable by depicting them in 3D environments [11]. However, according to [12], this field is still relatively unexplored and has two main challenges. First, rendering large, popular architectures like ResNet50 is currently not feasible with existing tools, which often restrict CNNs to linear structures without splits or joints.…”

Applications of Deep Learning for Drug Discovery Systems with BigData

“…Deep learning has become the mainstream of machine learning, and with the expansion of its applications, technical problems from the perspective of real-world problem solving have also emerged, primarily the black box problem: neural network machinelearning in deep learning is a black box [130][131][132][133][134][135][136][137][138][139][140][141][142][143][144]. The learning results are reflected in the node weights, and the obtained regularities and models are not represented in a form that humans can directly understand.…”