Neural Network

The lab is the core of the product and the closest thing to a neural network playground for beginners. You pick a 2D or 3D dataset for classification or regression, choose which features feed the network, then add or remove hidden layers while training runs. Every change redraws the neural network visualization, so a bad learning rate turns into a visibly unstable boundary rather than an abstract warning. Nothing you do here costs money or breaks an experiment.

Instead of a video followed by a quiz, the app sets tasks and lets you fail at them. A mission might ask you to separate two spirals with the fewest neurons possible, then grade the result against a target. The game framing keeps a session going past the point where a plain chapter would lose you, and each challenge maps to a concept that shows up later in the tutorial track.

The neural network tutorial library is ordered, not a pile of articles. It opens with what a network is and how a single neuron works, moves into how a network approximates an arbitrary function, then covers construction, training, and the problems that show up in practice. From there it branches into a CNN tutorial on convolutions and pooling, plus a recurrent network track, with separate chapters for 1D and 2D cases, pooling, seq2seq, and autoencoders.

Nine activation functions get their own pages: Linear, Sigmoid, Tanh, ReLU, ReLU6, LeakyReLU, ELU, Softplus, and Softmax. Each page pairs the formula with an interactive chart, so you can see where the gradient flattens and why a saturated Sigmoid slows learning. This section works as a standalone reference when you're debugging a model and can't remember which function fits the output layer.

The editor lets you assemble a network by dragging layers and setting input types, then computes each step as you go. Errors such as a mismatched input shape or an incompatible parameter show up immediately, which removes the usual cycle of train, crash, read traceback, guess. For someone who has never built a model, that instant feedback is the difference between an afternoon of confusion and a working first network.

The web version opens in any modern browser with nothing to install, which makes it the easy entry point on a borrowed laptop or a lab machine. The iOS and Android apps carry the same lessons plus offline access, and the iOS build also runs on Apple silicon Macs. Progress and content stay consistent across both, so you can start on a desktop and keep going on a train.