ML//neural network//dropout

- Randomly zero out neurons during training (typically 10-50%)

Randomly zero out neurons during training (typically 10-50%)

Forces redundancy — no single neuron can be essential.

Like training an ensemble of thinner networks that share weights.