A 'Brief' History of Neural Nets and Deep Learning, Part 4 (2015)

[sytelus]

http://www.andreykurenkov.com/writing/a-brief-history-of-neural-nets-and-deep-learning-part-4/

[andreyk]

Oh man, cool to see this on front page of HN again. For convenience, here is the first part where its best to start: http://www.andreykurenkov.com/writing/a-brief-history-of-neu...

One of these days I'll finally get around to making a PDF version that's a bit more convenient to save for lengthy reading and not just a series of blog posts...

[sytelus]

If you ever asked yourself, what did suddenly changed the importance of neural networks (ahm... deep learning), this is a great run down. The turning point in my view was 2010 paper by Schmidhuber where he conclusively proves that deeper networks have been learning all the time and excelling compared to SVM/k-means. They were just too slow so needed to run for 24+ hours. GPU made this 40X faster and suddenly deep networks were getting trained in couple of hours.

[woliveirajr]

Mods: it is from 2015, and this is the 4th part of the series.

[sctb → woliveirajr]

Thanks, we've updated the title.

[tomxor → woliveirajr]

Regardless, I've never seen it before and am glad it was posted. Not everything has to be released in the last 24hrs to be of interest here.

[CardenB → tomxor]

Yeah but it’s helpful to tag the headline with a date

[swsieber → CardenB]

Not only that, it's the convention. Here, a lack of a date implies a current date.