well then.
drxiv (12:17:22 AM): our project is actually kinda cool
drxiv (12:18:06 AM): we're learning statistical models of complex high dimensional nonlinear systems for use in bayesian action inference
drxiv (12:18:13 AM): okay so that was the unreadable version of what we're doing, haha
drxiv (12:18:33 AM): but the more easier to understand version is:
drxiv (12:18:38 AM): we have some model of a human arm
drxiv (12:18:45 AM): with muscles
drxiv (12:18:55 AM): and you have to figure out how to stimulate the muscles to make the hand move to a certain position
drxiv (12:19:07 AM): and the way the muscles effect the arm is fairly complex
drxiv (12:19:34 AM): so we're learning a statistical (non-equation) model of the arm, and using it to predict what to do using probabilities
drxiv (12:19:43 AM): and over time the system can learn how to move better and better
drxiv (12:20:10 AM): so for a robot it can learn how to move its arm better and better over time to reach certain objects in its field of view
drxiv (12:20:11 AM): and things like that
drxiv (12:20:15 AM): so its pretty cool :-)
nuggetynugg (12:20:23 AM): and um there are pretty colors involved, right?
drxiv (12:20:33 AM): well that graph was sort of just eye candy...
drxiv (12:21:09 AM): that just showed the entropy (certainty) of state transitions for areas around the training runs was higher than of other areas...
drxiv (12:21:23 AM): ie, where you've already been you're more likely to be able to know whats going to happen
drxiv (12:21:28 AM): rather than state you've never been in
drxiv (12:21:42 AM): but yeah, pretty colors :-)
nuggetynugg (12:21:46 AM): ok just checking.
drxiv (12:18:06 AM): we're learning statistical models of complex high dimensional nonlinear systems for use in bayesian action inference
drxiv (12:18:13 AM): okay so that was the unreadable version of what we're doing, haha
drxiv (12:18:33 AM): but the more easier to understand version is:
drxiv (12:18:38 AM): we have some model of a human arm
drxiv (12:18:45 AM): with muscles
drxiv (12:18:55 AM): and you have to figure out how to stimulate the muscles to make the hand move to a certain position
drxiv (12:19:07 AM): and the way the muscles effect the arm is fairly complex
drxiv (12:19:34 AM): so we're learning a statistical (non-equation) model of the arm, and using it to predict what to do using probabilities
drxiv (12:19:43 AM): and over time the system can learn how to move better and better
drxiv (12:20:10 AM): so for a robot it can learn how to move its arm better and better over time to reach certain objects in its field of view
drxiv (12:20:11 AM): and things like that
drxiv (12:20:15 AM): so its pretty cool :-)
nuggetynugg (12:20:23 AM): and um there are pretty colors involved, right?
drxiv (12:20:33 AM): well that graph was sort of just eye candy...
drxiv (12:21:09 AM): that just showed the entropy (certainty) of state transitions for areas around the training runs was higher than of other areas...
drxiv (12:21:23 AM): ie, where you've already been you're more likely to be able to know whats going to happen
drxiv (12:21:28 AM): rather than state you've never been in
drxiv (12:21:42 AM): but yeah, pretty colors :-)
nuggetynugg (12:21:46 AM): ok just checking.