ClassificationSample
Neural network (MLP)
Layers of simple neurons trained by gradient descent, able to learn curved boundaries.
Synthetic data, generated and trained in your browser.
Try this
- 01Watch the boundary form while training runs, and the loss curve fall.
- 02Set 1 hidden neuron on Circles: a single straight cut cannot enclose the inner ring.
- 03Switch to ReLU and press New weights a few times: the boundary is built from straight segments.
How it works
- 1Each hidden neuron computes a weighted sum of the inputs and applies a non-linear activation.
- 2The output neuron combines the hidden neurons into a probability.
- 3Backpropagation computes how each weight affects the loss.
- 4Gradient descent nudges every weight, epoch after epoch.
What the controls teach
- Hidden neurons
- Number of hidden neurons: each adds a direction the boundary can bend along.
- Activation
- The non-linearity. tanh gives smooth curves; ReLU gives piecewise-straight ones.
- Learning rate
- Step size of each update. Too high and the loss bounces; too low and training crawls.
Use it when
- Complex, non-linear patterns with enough data
Avoid it when
- Small data
- When individual predictions must be explained