I see lots of people stating unequivocally that the deep learning boom started with Krizhevsky et al 2012. But I see little credit given to Ciresan et al, who were winning image classification competitions in 2011 with deep convnets implemented in CUDA, trained on NVIDIA GPUs
-
Show this thread
-
Replying to @fchollet
Convnets on GPU predate CUDA: https://hal.inria.fr/inria-00112631/document …
3 replies 11 retweets 81 likes -
Replying to @goodfellow_ian
Which is the bigger breakthrough, fast convnets on GPU, or the knowledge that they can win hotly contested image classification competitions?
3 replies 7 retweets 24 likes -
Replying to @fchollet
I also won an ML contest in 2011 with conv nets on GPU (NIPS transfer learning contest) but I’m not claiming that was the start of the deep learning boom
1 reply 1 retweet 36 likes -
Replying to @goodfellow_ian
I don't think anyone is underestimating the significance of AlexNet in kick-starting the boom (obviously). But it's hard to deny that it was part of a trend at the time. Credit attribution is complex, and our community does not always do a great job at it.
1 reply 3 retweets 27 likes
More precisely: credit attribution tends to follow a winner-takes-all model, whereas innovation is more of an accretion process, where macro trends may well matter more than individual contributions.
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.