Uncertainty Quantification for Online Learning via Hierarchical Incremental Gradient Descent
Talk
Go to NumFOCUS academy page.
The Stochastic Gradient Descent algorithm is often used for online, large-scale machine learning problems but suffers from randomness in prediction. Weijie J. Su and Yuancheng Zhu provide a solution with their novel HiGrad algorithm that applies statistical inference to SGD predictions. This talk will explain the mechanics of HiGrad and demonstrate its implementation in Python.