Student Seminar
Abstract:
Many important biological processes are diffusion limited reactions
amenable to simulation by Brownian dynamics. The established methods
involve simulation of very many Brownian trajectories in a multidimensional
configuration space (CS); but the utility of such approaches is hampered
by the infrequent passage over entropic barriers. Computational resources
are wasted tracking "uninteresting" events.
The Weighted Ensemble Brownian (WEB) Dynamics method features a weighted
ensemble of trajectories in CS. Free energy barriers are surmounted
rapidly by splitting packets of probability until a minute portion
crosses the barrier. The new method is in excellent agreement with
established methods, but with far fewer (typically 3 to 4 orders of
magnitude) force calculations. As these dominate CPU usage, significant
speedups are attained in the simulation of diffusion limited processes.
Examples ranging from antibody-antigen binding to protein folding will
be presented.
The Department of Computer Science Seminar
Abstract: The partially observed Markov decision process (POMDP) is a model of sequential decision making under uncertainty that allows for state observations to be noise corrupted and/or costly. This added modeling validity, relative to the standard, completely observed MDP, is at the expense of decreased model tractability. New approaches for solving the POMDP will be presented that make use of a blend of various mathematical programming techniques and the genetic algorithm.
<--- 1997 Index