thanks for making the basics series, finding it very useful while doing the Andrew Ng courses!
@saifshaikh8679
19 күн бұрын
Please share notes with us
@mshonle
Ай бұрын
Hi, thanks for the video! I’ve a couple questions: What if one did try to use dropout during inference and not just training? Would that be a good way to get multiple, varied samples from a model? (e.g., instead of just using temperature for sampling) In training, does the order of the various batches matter? For example, after a few iterations and the weights start to settle, if you had a choice, would you want the next training examples to be the ones with the highest gradient differences or would you want to sharpen things first by picking the examples with lower perplexity? I anticipate the answer is going to be “random is better”?
Пікірлер: 3