GECCO-2012: Three EDA sessions, and more

Getting ready to start day 3 at GECCO-2012. Today, there are three sessions with EDA papers in them, so I hope to come to all of these. I’m also giving my second presentation in one of the sessions on distance-based bias in model-directed optimization of additively decomposable problems. The basic idea is to mine models from prior runs of model-directed optimizers for information about the problem domain to introduce bias when solving new problem instances of similar type, so that these instances can be solved with increased speed, accuracy, and reliability. Essentially, this allows an optimizer to learn from solving a certain type of problems to get better at solving similar problems in future. In this specific approach, models from prior runs are combined with a distance metric based on problem-specific knowledge. For more information about this work, you can download MEDAL Reports 2012001 and 2012004.

Comments are closed.