Experimental Results

In this section we will present experimental results that compare the performance of our adaptive parallel search strategy with each fixed strategy used exclusively for all problem instances. In particular, we will verify the following hypotheses in this section:

• EUREKA's adaptive search techniques can be used to achieve speedup over a variety of applications, and can demonstrate improved results over using a fixed strategy for all problem instances.

• The adaptive search technique can employ training examples from multiple applications to improve overall performance. We will demonstrate this using testing and training examples combined from two application domains.

• The learning component of EUREKA is able to significantly outperform any of the tested fixed strategies in terms of predicting the best strategy for a given problem instance.

• In addition to effectively making one strategy choice for a new problem, EUREKA is most effective of all tested approaches at making all strategy decisions for a given problem instance.

• A variety of learning techniques can be used to assist in strategy selection and offer speedup over serial search, though performance will vary from one learning technique to another.