Efficient learning with robust gradient descent

Matthew J. Holland, Kazushi Ikeda

Minimizing the empirical risk is a popular training strategy, but for learning tasks where the data may be noisy or heavy-tailed, one may require many observations in order to gene…