"AI's Next Brain Wave"

InformationWeek (04/25/05) No. 1036, P. 45; Ricadela, Aaron

IBM's Watson Laboratory, Microsoft Research, Intel, and the Xerox Palo Alto Research Center (PARC) are all engaged in artificial intelligence research that could pave the way for computer systems capable of learning from their users as well as their surrounding environment. IBM Watson's neurocomputing project seeks to add complexity to the artificial neural network model by developing layered networks whose behavior is dictated by the biological properties of vertebrate nervous systems; the project involves figuring out how these multilayered networks can circumvent the problem of writing programs with advanced knowledge of all the unknown factors of each task they may be faced with, and a March paper demonstrated that such networks could avoid the superposition catastrophe problem through pattern recognition. Microsoft Research scientist Eric Horvitz is focusing on how computers can serve as a memory aid to people through projects such as MemoryLens, which employs software to analyze a user's activities and schedules to recognize important events that can later function as reference points to pinpoint information. Horvitz's team is also working on prototype "streaming intelligence" software for relaying AI rules to smart phones. Intel is exploring various applications for "machine learning" techniques, including a system that uses a "learning engine" to project what chips on a silicon wafer are of good quality, and user interfaces that can more proactively predict user requirements via statistical methods. PARC's user-interface group is developing software such as ScentHighlights, which helps users scan through information by selecting key sentences from a document, which relate to keywords entered by the user. PARC research fellow Stuart Card says the software illustrates an emergent type of interface that responds to things that capture a user's interest.

Its been awhile since anything about machine learning has come into the news.

Full Article: http://www.informationweek.com/showArticle.jhtml?articleID=161501161