When learning from complex data such as medical scans or video feed we encounter long training periods and large memory demands. This makes searching for the best set of hyperparameters a hard to nearly impossible task. Most researchers simply try a few sets of hyperparameters by trial and error or use a grid search at best. These hyperparameters are usually chosen by some intelligent guess based on prior knowledge. It is clear to see this is not the best practice. A more DATA DRIVEN approach required.

In this session, we will learn practical methods to search for the best set of hyperparameters under an arbitrary limit of resources (run time or machine cost) by selecting the correct optimization algorithm and training settings.