Posted by Moshe LooksCross-posted with Open Source at Google blogTraditional machine learning systems work with relatively flat, uniform data representations, such as feature vectors, time-series, and probabilistic context-free grammars. However, reality often presents us with data which are best understood in terms of relations, types, hierarchies, and complex functional forms. The best representational scheme we computer scientists have for coping with this sort of complexity is computer programs. Yet there are comparatively few machine learning...
Monday, 10 November 2008
Subscribe to:
Posts (Atom)
