Move Over, Einstein: Machines Will Take It From Here
New Scientist (03/22/11) Justin Mullins
Cornell University Ph.D. student Michael Schmidt and professor Hod Lipson have developed a research technique that reverses the usual scientific method. The researchers first perform experiments and feed the results into a computer to discover the laws of nature, rather than starting with a hypothesis to test. The success of this method hinges on evolutionary computing, in which robots or computers are presented with a goal and then generate programs and combine the most promising ones until the objective is achieved. Through evolutionary computing, machines can execute tasks that they have not been programmed to do, and the technique is already being employed to address a host of problems ranging from aircraft design to creating train timetables. Schmidt and Lipson's evolutionary algorithm, Eureqa, has been automated and released for free online. One key to the algorithm's success is its ability to look for invariance in the equations it produces. Lipson thinks that Eureqa will enable the extraction of laws of nature from data at currently unheard of rates, and that this type of machine learning will become routine.
View Full Article - May Require Free Registration
No comments:
Post a Comment