Novel Algorithms Automatically Generated for Optimization Problems

A difficult task in computer science is designing algorithms. This task is particularly complex when efficient algorithms must be constructed for computationally difficult optimization problems. Two fundamental problems in both combinatorial optimization and machine learning are the maximum independent set problem and the automatic data clustering. The best specific algorithms for both problems are heuristic and have been constructed by combining primary heuristics. However, the possible combinations explored so far are a minimum number of the entire combinatorial space. The automatic exploration of such space would, therefore, allow finding more efficient algorithmic combinations. This article reports new algorithms for both problems constructed by the automatic generation of algorithms, an emerging field that allows to automatically produce an adequate algorithm for a specific set of instances of the problem. The best algorithms generated after the computational experimentation are reported and compared with existing heuristics, demonstrating that the new algorithms are competitive.