#1




project Sean Burke
mid point summary
Last edited by sburke; 04132010 at 09:04 AM. Reason: did not accept .odt extension 
#2




As we discussed, I'm a little worried here. For a "hill climbing" method to work, there has to be some notion of continuity. It's not a (useful) Monte Carlo method to just be guessing point solutions in a huge space in the hope of finding the right one. Please think carefully, for the problems that you propose, about what is the nature of the solution space, does uphill stepping make sense, etc.

#3




Final Project
Multiple minima (maxima) problem

#4




Interesting!
In slide 12, how do you actually evaluate the proposed functional exp(t d^2/dx^2) on a function? My guess is that you do it in Fourier space, where it is just a Gaussian lowpass filter. Why is that optimal? Then, in your slide 21, this would also correspond to the integral method that convolves with a Gaussian. In branch and bound (in this context) I'm not sure what is the class of problems for which you can know how to compute the bounds. 
Thread Tools  
Display Modes  

