CS-E4740 Gradient Methods
This lecture gives a glimpse on gradient methods that allow to tune or learn model parameters in ML methods. Gradient methods are based on a simple idea: given a current choice for model parameters, try to locally approximate the empirical risk by a linear function which is then minimized to obtain updated model parameters.
This lecture gives a glimpse on gradient methods that allow to tune or learn model parameters in ML methods. Gradient methods are based on a simple idea: given a current choice for model parameters, try to locally approximate the empirical risk by a linear function which is then minimized to obtain updated model parameters.