http://metamarkets.com/2011/machine-learning-in-wonderland/ Mon, 20 May 2013 23:58:35 +0000 hourly 1 http://wordpress.org/?v=3.5.1 http://metamarkets.com/2011/machine-learning-in-wonderland/#comment-21 Soeren Tue, 06 Mar 2012 12:39:57 +0000 http://metamarketsgroup.com/blog/?p=53#comment-21 "I get pitched regularly by startups doing "generic machine learning" which is, in all honesty, a pretty ridiculous idea." --- I disagree with such a statement. Standard off-the-shelf machine learning tools get you very far. In fact, the Netflix example that you have given is the perfect example for this. Unlike you claim, using only standard matrix factorization without any domain knowledge or data insight you could improve the RMSE already by 6%. Just one standard tool. No parameter tuning whatsoever lets you beat the cinematch score easily. Of course if you want to go further and squeeze the last tiny bit out then you need more powerful ideas and domain knowledge. I think the key to success is to do things right and in a clean and smart way. And if you need that last tiny bit, then of course you need expert knowledge.

]]> http://metamarkets.com/2011/machine-learning-in-wonderland/#comment-20 Phoenix Mon, 25 Apr 2011 15:09:00 +0000 http://metamarketsgroup.com/blog/?p=53#comment-20 As machine learning is gaining pace in the industry these insights are really enlightening. I have worked on machine learning earlier and agree completely with you on the fact that machine learning techniques should not be considered generic.

]]> http://metamarkets.com/2011/machine-learning-in-wonderland/#comment-19 Antonio Piccolboni Fri, 15 Apr 2011 19:44:18 +0000 http://metamarketsgroup.com/blog/?p=53#comment-19 For the opposite point of view watch Andrew Ng (https://www.youtube.com/watch?v=ZmNOAtZIgIk). I think it would be wise to suspend judgement as to how general ML will ever be, while acknowledging current limitations. Some people like to run startups and some like to solve big fundamental problems, let's all live in harmony.

]]> http://metamarkets.com/2011/machine-learning-in-wonderland/#comment-18 Foursquare’s Google Moment: Recommendations Launch Tonight Mon, 21 Mar 2011 23:31:42 +0000 http://metamarketsgroup.com/blog/?p=53#comment-18 [...] learning,” wrote Joseph Reisinger in a recent blog post titled Why Generic Machine Learning Fails, “is not undifferentiated heavy lifting, it’s not commoditizable like EC2, and closer [...]

]]> http://metamarkets.com/2011/machine-learning-in-wonderland/#comment-17 Idea 1: Data are caused | Navia Blog Mon, 21 Mar 2011 17:48:20 +0000 http://metamarketsgroup.com/blog/?p=53#comment-17 [...] put this in the context of a recent discussion: generic machine learning is, indeed, a false idol. Domain knowledge really is necessary to achieve an adequate explanation of many kinds of data. [...]

]]>