1 citations · 1 across the 4 of their papers we have counts for
1 paper · 2 filters
Max Lovig
In modern parametric model training, full-batch gradient descent (and its variants) suffers due to progressively stronger biasing towards the exact realization of training data; th…