Hod Lipson on The Robotic Scientist: Mining Experimental Data
TVO show "Big Ideas" presents Hod Lipson of Cornell University speaking about his amazing working in using evolutionary algorithms.
Through trial and error, robots learn how to walk (@ 9:49). This was achieve by give the robot basic motor commands, then the robot learns how to walk by try every combination of possible movements until it moves forward reaching its goal.
This same technique was used for evolutionary equations, by giving basic math primitives (+,-,/,*, sin,cos, etc) equations are built to match and model a best fit to an observable outcome. Hod calls this Symbolic Regression, and through millions of trials, a perfect and sometimes previously unknown equation will emerge!
Hod also points out, that trials with the most disagreement (@ 11:38) yields the most information! A new perspective the next time you get into an argument;)
This has huge ramifications for in area of robotics, programmable self-assemblies and AI. See my article on the prediction an AI God (a multi-consciousness AI), including Steven Hawking's (the smartest man in world) thoughts on warnings against the rise of AI.
Lipson lecture is entitled "The Robotic Scientist: Mining Experimental Data for Scientific Laws, from Cognitive Robots to Computational Biology."
(@ 21:43 section, using math primitives (+,-,/,*) this will evolve and find fitting equations)
http://creativemachines.cornell.edu/eureqa - A program for you to use to try evolutionary approach and was open source (now gone).
TVO show "Big Ideas" presents Hod Lipson of Cornell University speaking about his amazing working in using evolutionary algorithms.
Through trial and error, robots learn how to walk (@ 9:49). This was achieve by give the robot basic motor commands, then the robot learns how to walk by try every combination of possible movements until it moves forward reaching its goal.
This same technique was used for evolutionary equations, by giving basic math primitives (+,-,/,*, sin,cos, etc) equations are built to match and model a best fit to an observable outcome. Hod calls this Symbolic Regression, and through millions of trials, a perfect and sometimes previously unknown equation will emerge!
Hod also points out, that trials with the most disagreement (@ 11:38) yields the most information! A new perspective the next time you get into an argument;)
This has huge ramifications for in area of robotics, programmable self-assemblies and AI. See my article on the prediction an AI God (a multi-consciousness AI), including Steven Hawking's (the smartest man in world) thoughts on warnings against the rise of AI.
Lipson lecture is entitled "The Robotic Scientist: Mining Experimental Data for Scientific Laws, from Cognitive Robots to Computational Biology."
(@ 21:43 section, using math primitives (+,-,/,*) this will evolve and find fitting equations)
http://creativemachines.cornell.edu/eureqa - A program for you to use to try evolutionary approach and was open source (now gone).
No comments:
Post a Comment