Training neural network
Neural networks are trained directly in the browser.
The training data set contains several tens of thousands of handwritten digit samples mixed randomly.
The initial parameters you set affect the learning outcome.

Number of neurons.
This parameter determines the number of neurons in the hidden layer.
The more neurons there are, the longer the network takes to learn.
The minimum value is 4, the maximum value is 32.

Learning rate.
The multiplier of changes when the training function converges to the minimum of errors.
The minimum value is 0.01, the maximum value is 1.

Start error threshold.
If the error of the random initial network state exceeds the specified value, the network will be initialized again.
The minimum value is 0.01, the maximum value is 1.

Stop error threshold.
If the difference between the errors of the training iterations is less than the specified threshold, the training is terminated.
The minimum value is 0.0002, the maximum value is 0.01.

Min / max initial weight.
Boundary values when forming the initial state of neural network connections.

Hidden / output layer bias.
Offset of the result of the neuron activation function.
Trained networks can be saved either locally or remotely to the server (if you have access rights).
Locally stored networks are only available in the current browser.
Remote networks are available to all users.