tanh.xyz - Neural Network Playground

Train a neural network to classify two interleaving half-moon shapes. Watch the decision boundary evolve in real-time as the network learns.

Online clustering algorithms with bounded cluster sizes. Explore different data domains and visualizations.

Online MNIST training with seeded reproducibility, live train/test curves, per-digit recall, and top gradient-norm samples.

Train a compact model on 10-digit addition and watch the grokking-style train/test loss transition across curriculum phases.

Interactive simulation of a linear dynamical system over time. Observe state evolution and matrix operations.

Explore how autoencoders compress and reconstruct data. Visualize the latent space and see how the network learns efficient representations.

Visualize different optimization algorithms navigating loss landscapes. Compare SGD, Adam, RMSprop, and more on various terrain types.