 |
Description
In difficult real-world learning tasks such as controlling robots, playing
games or pursuing and evading an enemy, there are no direct targets that
would specify correct actions for each situation. In such problems, optimal
behavior must be learned by exploring different actions and assigning
credit for good decisions based on sparse reinforcement feedback. Our
research in this area focuses on methods for evolving Neural Networks
with Genetic Algorithms, i.e. Evolutionary Reinforcement Learning, or
Neuroevolution. Compared to the standard Reinforcement Learning, Neuroevolution
is often more robust against noisy and incomplete input, and allows representing
continuous states and actions naturally. Our methods include utilizing
subpopulations, population statistics and knowledge in the population
and evolving network structure. Much of this research involves comparisons
of neuroevolution to traditional methods in several benchmark tasks such
as pole balancing and mobile robot control. In addition, we have applied
the techniques to a variety of domains, including robot control, game
playing, resource optimization, music generation, theorem proving and
modeling language evolution.
|
|
|
Researchers
|
 |
Matt Alden, Joe Bruce, Bobby Bryant, Tino Gomez, Paul McQuesten, Lisa
Redford, Ken Stanley, Shimon Whiteson
|
|
