Our work estimates gradients over multiresolutional regions (learned by clustering) using previously evaluated function calls in order to solve black box global optimization problems. This approach is best suited for problems where it is expensive/time-consuming to evaluate functions and/or for high-dimensional problems.

Personnel:

Maya R. Gupta (EE Associate Professor)

Megan Hazen (APL Fellow)

Publications:

"A Quasi EM Method for Estimating Multiple Transmitter Locations," Jill K. Nelson, Jaime Almodovar, Maya R. Gupta and Will Mortensen, IEEE Signal Processing Letters, 2009.

"Estimating Multiple Transmitter Locations from Power Measurements at Multiple Receivers," Jill K. Nelson, Jaime Almodovar, Maya Gupta, William Mortensen, Proc. ICASSP, 2009.

"Gradient Estimation in Global Optimization Algorithms," Megan Hazen and Maya R. Gupta, Congress on Evolutionary Computation (to appear), 2009.

"A multiresolutional estimated gradient architecture for global optimization," Megan Hazen and Maya R. Gupta, Congress on Evolutionary Computing , 2006.

"Global Optimization for Multiple Transmitter Localization," J. K. Nelson, M. U. Hazen and M. R. Gupta, MILCOM, 2006.

Related Work:

Similarity-based Learning