2 ms·

That still looks like a “converge faster” paper. https://arxiv.org/abs/2006.10732 https://arxiv.org/abs/2006.10732 The above provides a nuanced theoretical vi

by vladf 7mo ago

That still looks like a “converge faster” paper. https://arxiv.org/abs/2006.10732 https://arxiv.org/abs/2006.10732 The above provides a nuanced theoretical view. GD inductive bias is probably better unless your model is misspecified