Path: utzoo!attcan!uunet!cs.utexas.edu!wuarchive!cec2!news From: ddj7203@cec1.wustl.edu (David D. Jensen) Newsgroups: comp.ai Subject: Re: Re: The AI Breakthough -- What It Will Be Like !!! Message-ID: <1990Nov5.145716.19427@cec1.wustl.edu> Date: 5 Nov 90 14:57:16 GMT References: <1870006@hpwrce.HP.COM> Reply-To: ddj7203@cec1.wustl.edu (David D. Jensen) Organization: Washington University, St. Louis MO Lines: 36 In article <1870006@hpwrce.HP.COM> kingsley@hpwrce.HP.COM (Kingsley Morse) writes: > OPPORTUNITY > >If CLASSIFICATION ALGORITHMs that don't use a lot of CPU on large problems >exist, they could be applied widely in challenging areas such as speech >recognition, forecasting the stock market, forecasting weather, artificial >intelligence, pattern recognition, neural nets, etc..... Perhaps you're working >on one. See the end of this posting for a detailed description of this type >of problem. > ... >I propose we use this entry on the net as a grass roots "CLEARINGHOUSE" for >people interested in such algorithms. For example, If you are working with an >algorithm which classifies a LOT of data, you can tell others about it here; >the algorithm's name and computational complexity for large problems would >be helpful. More than just computational complexity is important here. We should also consider: - Performance: How well does the algorithm classify new cases? This is often a largely empirical question because we lack good answers to the question: "How is the world really structured?". Consequently, we often just take a pool of (what we think are) representative problems and test the various algorithms. - Use of prior knowledge: What if we already know something about the problem? Can this knowledge be used by the algorithm? Must the knowledge be explicitly encoded, or is more flexible human guidance a possibility? - Representation of results: How are the results expressed? Are these results directly usable by other programs? Are the results directly usable by humans? How much do they aid our further exploration of the problem? My overall point is: Don't confuse "fast" with "useful".