101 - The lottery ticket hypothesis, with Jonathan Frankle
NLP Highlights
00:00
How to Train a Lottery Ticket Faster?
For amnistan c far ten, we found that running this lottery ticket experiment, we'll find very small sub networks that match the performance original network. We got sparsey levels that were on par with what you would get after pruning the network or pruning th network after training. And we found that these networks matched the accuracy of the original network. In fact, sometimes they exceeded the accuracy ofThe test accuracy, and they learned faster. That is to say, iteration for iteration, they reached higher accuracy than the full network had reached.
Transcript
Play full episode
