Active Mini-Batch Sampling using Repulsive Point Processes

by · Jun 14, 2019 · 128 views ·

ICML 2019

We explore active mini-batch selection using repulsive point processes for stochastic gradient descent (SGD). Our approach simultaneously introduces active bias and leads to stochastic gradients with lower variance. We show theoretically and empirically that our approach improves over standard SGD both in terms of convergence speed as well as final model performance.