« Odyssey Rosso | Main | 北村(2000) »
June 10, 2005
Pedro(1991)
Pedro S. de Souza, Sorosh N. Talukdar. "Genetic Algorithms in
Asynchronous Team". Proceedings of the Fourth International Conference
on Genetic Algorithms, pp.392-397, 1991.
GA関連サーベイの続き．
GAとNewton法のそれぞれの強みを活かして，探索を2つのアプローチを組み合わせることで行っていこう，という論文．
グローバルな探索をGAでざっくりと行って，Newton法とかでローカルな探索をぐりぐりとやっていくと．
A GA could be applied to the entire set of equations, but
it would be very slow. Instead, we apply a GA only to the global
variables and use conventional algorithms (Newton's Method) for the
local variables.
Pedro(1991) pp.395
チームベース(中央管理の不在)で非同期的に獲得した情報の交換を行うことを実現している．
なかなか参考になる．
To emulate the organizations of social insects in
software, we make two modifications to the structure of a supervised
group or blackboard (Fig. 1). First, we eliminate the supervisor.
Second, we eliminate the need for supervision by encapsulating each
member in a small, rule-based program that gives it consensus-building
capabilities. The result is an unsupervised group of the form shown in
Fig. 2. We call this structure an A-Team to emphasize that members
interacts asynchronously, that is, they exchange information when they
find it convenient to do so, rather than at predetermined points in
time. These exchanges occurs through addtions to, or modifications of,
the contents of the shared memory.
Pedro(1991) pp.393
基本的にはGAでの計算量といったら，評価回数ということらしい．
で，解を見つけるまでのコスト(計算量)が減りましたよと．
We have demonstrated that genetic algorithms can be
combined with other algorithms to form A-teams for solving sets of
nonlinear algebraic equations. These teams result in considerable
savings in the amount of computational effort (measured in terms of
function evaluations) required to find solutions. In the combinations
we have discussed, the genetic algorithms provide starting values for
less robust but faster algorithms.
Pedro(1991) pp.396
Posted by ysk5 at June 10, 2005 10:42 AM
Trackback Pings
TrackBack URL for this entry:
http://ysk5.s58.xrea.com/mt/mt-tb.cgi/616