For my purposes I would not initially parse the scene directly into a tree but would expand the target structure set from one = tree to other structures. Though tree is generally isomorphic to many other structures there might be more of an immediate mapping based on scene content. Though typically papers like this are more concerned with direct usage for example as in this case, scene classification. Also this paper is oriented towards NLP so tree is appropriate here I suppose.
John From: YKY (Yan King Yin, 甄景贤) [mailto:[email protected]] Sent: Saturday, February 8, 2014 3:52 AM To: AGI Subject: Re: [agi] Ben's geometry of mind paper On second thought, about this paper: "Parsing Natural Scenes and Natural Language with Recursive Neural Networks" Socher, Lin, Ng, Manning (2011) They have pre-processed the image scenes so that each image contains less than 100 features. That is really "small data" and a logic engine would have no problem constructing a parse tree of the image using a bunch of first-order rules. I guess the research code's speed would be comparable to that of a logic engine's, if not slower... but that is regarding recognition task. What is special about their method is that it can learn from data efficiently. The learning can also be done by a logic engine, but that would be much slower... AGI | <https://www.listbox.com/member/archive/303/=now> Archives <https://www.listbox.com/member/archive/rss/303/248029-3b178a58> | <https://www.listbox.com/member/?&;> Modify Your Subscription <http://www.listbox.com> ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
