1 / 9

Genetic Algorithms on Steroids Using a parallel approach to help evolution along...

Genetic Algorithms on Steroids Using a parallel approach to help evolution along. Asanka Herath & Buddhika Kottahachchi. Motivation. Genetic Algorithm's are generally slow. Constrained by the dependence on randomness to create good genes.

zandra
Télécharger la présentation

Genetic Algorithms on Steroids Using a parallel approach to help evolution along...

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Genetic Algorithms on SteroidsUsing a parallel approach to help evolution along... Asanka Herath & Buddhika Kottahachchi

  2. Motivation • Genetic Algorithm's are generally slow. • Constrained by the dependence on randomness to create good genes. • Parallelizing usually just allows the examination of a larger set of candidates – still constrained by randomness to provide good genes.

  3. What if? • Use a different randomized but faster algorithm to generate candidate solutions in parallel. • Inject these candidate solutions into the gene pool of the Genetic Algorithm. • Help the fitness level of the gene pool improve faster than with pure randomization... ... leads to better solutions quicker?

  4. Problem Domain: Bin Packing Rectangles • Given a set of rectangles and the width of a bin, determine the minimum height for a bin containing those rectangles. • Only allow two orientations (ie. Horizontal & Vertical).

  5. Approach • 1 Node devoted as a Coordinator. • Half of the others run a fast simulated annealing algorithm (Candidate Generator). • The rest run independent GA's. • Coordinator polls for candidates and pushes them out to the GA's between GA iterations • Uses C++ and MPI

  6. Validation Method • 60 rectangles (side length up to 100 units)Hopper E. and Turton B. C. H., 2002, "An empirical study of meta-heuristics applied to 2D rectangular bin packing" Special Issue on Cutting, Packing and Knapsacking Problems, Studia Informatica, vol. 2, no. 1 • Bin width 100 units • 4 Nodes/16 virtual Nodes • 50/100 GA iterations, 50 SA iterations per GA iteration.

  7. Results • Control case runs all independent Genetic Algorithms • Test case injects “good genes” into the Genetic Algorithms

  8. Observations • Solutions generated are very close to optimal (~ 10% wastage) • Small improvements required vast amounts of computation • This approach on average yielded about .5% improvement in solution quality (based on wastage).

  9. Conclusions • Given the constraints – the improvement is significant. • Problem selected to test hypothesis – non-ideal? • This approach merits further investigation • Other problem domains

More Related