Learning rate in deep learning and most learning algorithms is a hyper-parameter that determines the step size at each iteration while moving towards a minimum of a loss function. Essentially, it controls how much we are adjusting the weights with the respect to the loss gradient. With learning rate values like 0.1 and 0.01 you mentioned, the model learns by adjusting the weights by 10% and 1% of the gradient, respectively. Therefore, learning rate is directly related to how fast or slow a network trains. However, it's important to set a careful balance for learning rate. If it's too large, the algorithm might overshoot the minimum and diverge. Conversely, if it's too small, the algorithm will need more iterations to converge to the minimum, meaning it will work slowly. A common practice is to gradually decrease the learning rate, allowing the model to settle more comfortably into the global (or local) minimum. Procedures for changing the learning rate during training are called learning schedule or learning rate schedule. Examples include: time-based decay, drop-based decay, etc. To find an optimal learning rate, consider using grid search or a learning rate finder tool that will try multiple learning rates and see which one provides the least loss.

In deep learning, the learning rate is a hyperparameter which determines how much we are adjusting the weights of our network with respect the loss gradient. Learning rate directly impacts how quickly our model can converge to a local or global minima (i.e., lower error). The numbers you see (like 0.1, 0.01 etc.) are the actual values of the learning rate, typically between 0 and 1. These values define the step size during gradient descent. A smaller learning rate requires more training epochs given the smaller updates made to the weights each step, but with large learning rates we may skip over optimal solutions, or even fail to converge. Choosing the right learning rate is crucial. Too high, the gradient descent possibly overshoots the minimum. Too low, gradient descent may be slow to converge or get stuck in an undesirable local minimum. Oftentimes, the learning rate is set via trial-and-error or more systematic approaches such as Grid Search, Random Search or adaptive methods. Adaptive methods like AdaGrad, RMSProp, or Adam can adjust the learning rate throughout the training, based on the gradients.