This repository is a collection of interesting visualization work (VIS) in the domain of artificial intelligence (AI). As AI is getting applied to more and more domains, even crtitical ones, it is important to be able to understand the decisions that a model might make.
As there is a lot of visualization work in the broad field of AI, the following subdivides this corpus into different application areas.
- Activation Atlas
- Building Blocks of Interpretability
- Circuits
- Feature Visualization
- OpenAI Microscope
This section shows the aforementioned tools in detail, providing a short explanation and helpful links.
Interactive article on how to bring feature visualization into context to extract a bigger picture.
Interactive article on how to make image classification more interpretable.
Interactive article on the reverse-engineering of neural network substructures.
Educational tool that shows the inner workings of a Convolutional Neural Network.
Live Demo | Code | Video | Paper
A tool for the interactive exploration of Convolutional Neural Networks (Convnets or CNNs).
Showing high-dimensional data as low-dimensional embeddings.
Educational tool for learning about recurrent neural networks (RNNs).
Optimizing the input image through backpropagation to obtain representative images for neurons or neuron clusters.
Educational tool that lets users experiment with GANs and how to train them.
Automatically obtaining overview visualizations for neural network architectures directly from Keras code.
Feature Visualizations and other visualization techniques for many different architectures and neurons within them.
Latex plugin to draw neural network figures.
Interactive learning environment for training simple neural networks.
Using the trained classifier to find erroneously labeled samples in a dataset.
Interactive article that shows how a RNN stores information over multiple time steps.
