How Neural Networks Transform Shapes Built by Lijian Liu

Purpose: Visualize how neural networks transform 2D shapes. Page starts with auto-random mode updating weights/biases continuously. Learn: Weights control rotation/scale/shear, biases control translation, activations add non-linear warping, more neurons = more expressive power. Controls: Click any weight/bias value to auto-animate its effect, click again to pause, drag slider or type directly for manual control, check "Apply" to keep changes.