Title: Mixed Integer Nonlinear Optimization: Algorithms and Software.
Speaker: Dr. Ashutosh Mahajan, Mathematics and Computer Sc. Division,
Argonne National Laboratory, USA.
Abstract:
Mixed Integer Nonlinear Programming (MINLP) finds applications in a wide
variety of applications in science and engineering. These optimization
problems suffer from two difficulties: presence of integer variables that
are used to model on/off decisions or to represent discrete quantities, and
the possibly nonconvex nonlinear functions in objective and constraints. We
present Minotaur, a next-generation open-source solver for MINLP, to solve
these problems. Minotaur has been built ground-up for solving MINLP and
global optimization problems. It offers a variety of algorithms:
Branch-and-Bound, Quesada-Grossmann, Branch-and-Reduce for solving these
problems. In this talk, we will first describe briefly the design of
Minotaur toolkit, and compare the performance of its default algorithms
with that of existing state-of-the-art solvers. Next, we will describe
three new mathematical techniques that we have developed and implemented
in Minotaur: (i) Exploitation of hidden convexity in some nonconvex
quadratic constraints, (ii) Reformulation techniques for nonlinear
constraints, and (iii) Quadratic Programming (QP) approximation based
branch-and-bound algorithm. We will finish with some ideas for future
work.
Time and Date: 10AM, Monday, 10/10/2011 (in Skype mode)
Venue: Conference Room, First floor, Next to CSE Office, Kanwal Rekhi
Building
Speaker's Bio: Dr. Ashutosh Mahajan is presently a PostDoctoral
Appiontee with Mathematics and Computer Sc. Division, Argonne National
Laboratory, USA. His current academic interests are integer programming,
mixed integer nonlinear programming, mathematical programming, etc. He
has a PhD in Industrial Engineering from Lehigh University and a BTech
in Production and Industrial Engineering from IIT Delhi.
News Category