[Home]
[
News]
[ Events]
[ People]
[ Research]
[ Education]
[Visitor Info]
[UCSD Only]
[Admin]
Line Search Algorithms for Projected-Gradient Quasi-Newton Methods
Michael Ferry
UCSD
Abstract:
We briefly survey line search algorithms for unconstrained optimization. Next, we consider the search direction and line search strategies used in several algorithms that implement a quasi-Newton method for simple bounds, including algorithm L-BFGS-B. In this context, we discuss two currently-used line search algorithms and introduce a new method meant to combine the best properties of two different strategies. We present a modified L-BFGS-B method using the new line search and demonstrate its significant performance gains by numerical tests using the CUTEr test set.
Tuesday, November 10, 2009
11:00AM AP&M 2402
News]
[ Events]
[ People]
[ Research]
[ Education]
[Visitor Info]
[UCSD Only]
[Admin]
Directors:
Randolph E. Bank
Philip E. Gill
Michael Holst
Administrative Contact:
[ Click to Send Email ]
Randolph E. Bank
Philip E. Gill
Michael Holst
Administrative Contact:
[ Click to Send Email ]
Michael Ferry
UCSD
Abstract:
We briefly survey line search algorithms for unconstrained optimization. Next, we consider the search direction and line search strategies used in several algorithms that implement a quasi-Newton method for simple bounds, including algorithm L-BFGS-B. In this context, we discuss two currently-used line search algorithms and introduce a new method meant to combine the best properties of two different strategies. We present a modified L-BFGS-B method using the new line search and demonstrate its significant performance gains by numerical tests using the CUTEr test set.
Tuesday, November 10, 2009
11:00AM AP&M 2402