Which value is utilized in the Stochastic Gradient Descent algorithm to control the learning speed?

The learning rate is a crucial hyperparameter in the Stochastic Gradient Descent (SGD) algorithm because it determines how much to change the model's weights in response to the estimated error each time the model weights are updated. Specifically, the learning rate controls the size of the steps taken towards the minimum of the loss function. A higher learning rate can lead to faster convergence but may also risk overshooting the optimal solution. In contrast, a lower learning rate allows for finer updates, which can lead to better convergence but may slow down the training process.

Setting an appropriate learning rate is vital for effective training; if set too high, the optimization may diverge or oscillate, while a rate that is too low can lead to unnecessarily long training times and a risk of getting stuck in local minima. The learning rate, therefore, directly affects both the speed and stability of the convergence during the training process.