3-5 August 2022
Universität Klagenfurt
Europe/Vienna timezone

Using Statistics to Determine the Learning Rate for Gradient Descent

3 Aug 2022, 17:30
20m
HS 4 (Universität Klagenfurt)

HS 4

Universität Klagenfurt

Talk Statistics Session B2 Statistics

Description

While gradient descent is ubiquitous in Machine Learning, there is no adaptive way to select a learning rate yet. This forces practitioners to do "hyperparameter tuning". We review how optimization schemes can be motivated using Taylor approximations and develop intuition why this results in unknown hyperparameters. We then replace the Taylor approximation with a statistical Best Linear Unbiased Estimator (BLUE) and derive gradient descent again. But this time with calculable learning rates.

Primary author

Felix Benning (Universität Mannheim)

Presentation Materials

There are no materials yet.