Dynamical SystemsVisualizations
RNN
Inside the Model
Explore hidden state trajectories, gradient flow through time, eigenvalue spectra, and how individual neurons specialize. Inspect every parameter at every training step.
Inside the RNN visualizes a trained character-level RNN with real data. You can explore hidden state trajectories projected into PCA space, trace gradient flow through time to see vanishing effects, inspect the eigenvalue spectrum and spectral radius evolution, and observe how Elman-style categories emerge from prediction patterns. Memory diagnostics and weight norm evolution are also available.