Adaptive stochastic optimization via trajectory cues

Training machine learning models requires users to select many tuning parameters. For example, a popular training method, stochastic gradient descent, requires users to select a value for the learning rate. These tuning parameters are hard to select and users often resort to time consuming trial-and-error process to find a good set of parameters. This project aims to automate the selection of these tuning parameters by using information inferred from training algorithm trajectories. This will reduce the time spent training machine learning models, and make machine learning more user-friendly