Distributed Neural Networks with GPUs in the AWS Cloud
techblog.netflix.comIsn't this infringing the recent patent of Neurala? http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2=H...
Holy crap. Am I reading this correctly? Is this really a patent for running learning algorithms on the GPU?
I don't see how the patent office didn't just google for 30 seconds...
http://www.sciencedirect.com/science/article/pii/S0031320304...
Maybe the patent has interesting twists and turns, but applying known algorithms on previously known architectures doesn't seem like the innovation that they should hand out protection for 17 years.
The Netflix tech blog goes into more detail about running deep learning on Amazon's cloud servers.
http://techblog.netflix.com/2014/02/distributed-neural-netwo...
Is this really a good idea? I thought neural networks didn't do as well as other methods outside of image classification and problems like it. Many of the advancements in Deep Learning were learning to extract features from unlabeled data. I assume all of Netflix's data is labelled.
EDIT: The deaded comment below me makes a good point, but they still have labels in the sense of what everyone watched and for how long.
Also the new title is terrible. Oh I guess it's a different article now???
Feature extraction can be aided by unsupervised data but will certainly work with labeled data. One of the advancements bundled under 'deep learning' is how we can leverage unlabeled data (which is much easier to come by) to improve performance. And of course you can always do unsupervised learning with labeled data, just toss out the labels ;)
It's actually the multiple layers hidden units that perform non-linear feature extraction and the unsupervised pre-training is simply a means to do this better (theoretically, although we don't really know what's happening as much as it would seem).
Most of the current research shows deep neural nets to be state of the art in image classification and nlp. I don't know that it is the case that deep learning techniques do not work out side this area, it's just there hasn't been much published on it either way. Although I do believe the Kaggle Merck contest was neither of these, and deep learning out performed all other techniques http://blog.kaggle.com/2012/11/01/deep-learning-how-i-did-it...
Deep learning just refers to the idea of a many layered neural network. These can be used for unsupervised (unlabeled) learning and supervised learning (labelled).
Many of the recent headline results for deep learning have been supervised such as the image net classification challenge (http://www.image-net.org/challenges/LSVRC/2013/)
A mod changed the link and title to the netflix tech blog entry, which is far more appropriate for this audience anyway. The wired thing was sort of blogspam.
Has any large-scale genetic algorithm infrastructure been developed? I've often wondered if EaaS (Evolution as a Service) could ever function successfully. Surely it has potential.
I wish Netflix would just give me a list of genres I could browse through and stop with the oh-my-so-clever recommendation engine. Maybe I want to stretch my viewing habits... how am I going to see what is on offer if it is always filtered through what I chosen before?
I have made more than one good decision choosing from the Random Picks list. For all I know they're not random and very calculated, but they seem to 'stretch' my taste.
It has that at the top of the page
Yea, but it's harder to get the specific ones like 'gritty crime dramas from the 80's with strong female leads'.
The best possible side-effect from this? That we finally figure out a universal theory of film production and consumption. The social scientific/cultural sociological significance of this is quite lovely.
Sadly, the Recommendation Engine is only of limited use...and I hate to be one of those naysayers as I understand the economics and contractual issues here...the movies that Netflix thinks I would really enjoy and are available for streaming and I haven't seen (I watch a lot of movies but am by no means prolific) goes downhill after a couple dozen.
It used to be that on the iPad, at least two bottom rows were dedicated to "New Releases" and "Recently Added"...sometimes neither of those rows seem to show up, and so I find myself logging into the web client just to see those listings, and -- I assume this is why they aren't as spotlighted in the iPad app anymore -- there's generally not much new to see. While I like House of Cards, I think Netflix would appeal to me much better if it spent that money on 200 - 300 good recent releases, or holding steady on to some of the great classics (there used to be more Akira Kurosawa and Woody Allen movies on Instant).
I don't find the recommendations very useful either. Maybe the secret value of the recommendation engine for Netflix is evaluating movies outside of their catalog. When considering how much to pay for the streaming rights to a movie, they can use the recommendation engine to determine how well received it would be by their membership and bid accordingly.
Really a good news. Looking something like this
What is it's first recommendation? Add more DRM into the web standards? Subvert the open web further?