<aside> 👨🏫
Pedram Jahangiry, Ph.D., CFA | Associate Professor of Data Analytics & Information Systems | Huntsman School of Business, Utah State University
</aside>
<aside> 🧠
A curated collection of interactive neural network visualizations — organized from beginner to advanced. All tools are free, browser-based, and require no installation.
</aside>
BeginnerThe gold standard for neural network visualization. Built by Daniel Smilkov and Shan Carter at Google. Adjust hidden layers, neurons per layer, activation functions (ReLU, Tanh, Sigmoid, Linear), learning rate, and regularization — then watch the decision boundary evolve in real time on 2D classification datasets. Start here.
BeginnerA simple, focused visualization of backpropagation in action. Watch individual weights update in real time as the network learns. Great for building intuition about how gradient descent works at the connection level — one of the hardest concepts to grasp from equations alone.
IntermediateAn enhanced, modern fork of TensorFlow Playground with additional capabilities. Includes modern activations (Swish, ELU, Leaky ReLU, SoftPlus), dropout, momentum, layer-wise gradient normalization, and learning-rate autotuning. Same beautiful interface, but lets you experiment with deeper concepts without leaving the browser.
Advanceddistill.pub/2019/activation-atlas
Visually stunning. Uses feature inversion to visualize millions of activations from an image classification network, creating an explorable atlas of learned features. Shows what deep networks actually "see" at different layers. Best appreciated after understanding how convolutional and fully connected layers work.