Neural Network Playground
A TensorFlow-Playground-style multilayer perceptron trained entirely in the page by hand-written backpropagation. Pick a two-class dataset — two gaussian blobs, XOR quadrants, concentric circles or two interleaved spirals — set the label noise, then shape the network: one to three hidden layers of up to eight neurons each, with tanh, ReLU or sigmoid activations and your choice of learning rate. Press Play and minibatch stochastic gradient descent runs epoch after epoch while the canvas repaints the decision function as a blue/orange probability heatmap with the training points overlaid (test points drawn hollow). A second chart tracks train and test loss so you can watch the gap open when the model overfits, and Reset re-rolls the weights for a fresh start.
Runs 100% in your browser — models are trained and computed locally on your device.
Read the full guide to this tool
Notes
- The output neuron is a sigmoid trained on binary cross-entropy; hidden layers use the activation you pick and are updated by minibatch SGD with batch size 10 at the selected learning rate.
- A network with no nonlinear hidden layer can only draw a straight boundary, so XOR, circles and spirals are unlearnable until at least one hidden layer of nonlinear units bends it.
- Weights start from Xavier/He-style scaling — standard deviation √(1/fan-in), or √(2/fan-in) for ReLU; too high a learning rate makes the loss diverge instead of shrink.
- Runs 100% in your browser — models are trained and computed locally on your device.
Related tools
Backpropagation Step-by-Step
Walk through one forward and backward pass of a tiny neural network, number by number.
Handwritten Digit Recognizer
Draw a digit and watch a neural network trained on MNIST recognize it live in your browser.
Activation Function Explorer
Compare sigmoid, tanh, ReLU and friends with their derivatives, and see why deep gradients vanish.
Convolution Filter Visualizer
Slide a 3×3 kernel over an image and watch convolution, ReLU and max-pooling build the output pixel by pixel.
Hopfield Network
Store pixel patterns in a Hopfield memory, corrupt them, and watch the network recall the original.
Self-Attention Visualizer
See how Transformer tokens attend to each other with arcs, heatmaps, heads and causal masks.