HN Explorer

Ask HN: Why is Deep learning winning Genetic algorithms?

5 points by real-hacker ↗ HN
Both can be seen as strategies of searching for a solution in an optimization problem. Why does deep learning (gradient descent) work better? Is it because of the randomness in GA? And are there any problems more suitable for GA, where deep learning fails?

1 comment

[ 1.7 ms ] story [ 17.6 ms ] thread
opless
I feel that it's because of the initial hype. You get more eyes on the problem, which generally creates more interest as people publish their results. As much of the results (that get reported) are in computer vision that generates more interest from the less technical, and you get that critical mass where projects like Twitter and Facebook start growing simply because it's being talked about so much.

The GA and Symbolic fields seem to be languishing only because relatively speaking no one is talking about them. Also by now it's probably easier to get funding for "deep learning".