Massively parallel systems and global optimisation
Speaker: DR. NARENDRA KARMARKAR , Laboratory for Computational Mathematics
Date: July 25 2008
Time: 12:30PM to 1:30PM
Location: 4-237
Contact: Shirley A. Entzminger, 617-253-4357, daisymae@math.mit.edu
Relevant URL:
*******************************************************************************
COMPUTATIONAL RESEARCH in BOSTON SEMINAR
[NOTE: Different location.]
DATE: Friday, JULY 25, 2008
TIME: 12:30 PM
LOCATION: Building 4, Room 237
(Pizza and beverages will be provided.)
Title:MASSIVELY PARALLEL SYSTEMS AND GLOBAL OPTIMISATION
Speaker:DR. NARENDRA KARMARKAR
Laboratory for Computational Mathematics
ABSTRACT:
We will briefly describe recent breakthrough in design of massively parallel systems based on insights derived from global optimization problems having multiple global optima. These designs include:
- Physical design of the projective geometry machine using massively
parallel quantum tunneling, which can totally overcome obstacles of
latency and bandwidth faced by contemporary designs. The new design can
broaden applicability of massive multi-threading to large and very
general classes of computational problems, and can be implemented using
already known fabrication techniques.
- Design of multi-ported, low latency, secondary storage based on
magneto-optics, implementing shared memory directly at physical level,
providing a highly valuable feature for data bases and transactional
memory.
- Design of new high bandwidth switches required for next generation
internet infrastructure.
- Design of novel robots with large number of "electro-magnetic
fingers" for placing atoms based on complex and sparse patterns of
multiple global minima that are more general than regular periodic
patterns achieved before using interference lithography.
- Design of control systems whose stability analysis requires
liapunov-like functions with multiple basins of attraction.
- Design of phased-array radars in tera-hertz range.
- Computational calibration of parameters occurring in empirical
force fields, whose values may be difficult to measure experimentally
but can be reverse engineered from known structure of folded proteins.
This work involves integration of ideas, concepts and processes from many fields:
- Math (Optimization theory, Discrete subgroups of lie groups)
- CS (Parallel Architectures)
- Physics (Path integrals, Quantum tunneling, Optics, Electron Optics)
- EE (CMOS & MEMS processes, Field emission devices, Control Theory)
- Material Science (Generalized interference lithography)
[NOTE: "Extended Abstract" is attached to e-mail and can be found on the CRiB websit at the end of end of Dr. KARMARKAR's abstract.]
*******************************************************************************
Massachusetts Institute of Technology
Cambridge, MA 02139
http://www-math. mit.edu/crib
For information on CRiB, contact:
Alan Edelman: edelman@math.mit.edu
Steven G. Johnson: stevenj@math.mit.edu
Jeremy Kepner: kepner@ll.mit.edu
Patrick Dreher: dreher@mit.edu
See other events that are part of
See other events happening in July 2008
