Keeping in line with our fascination of technical ideas, we would like to share another of our research proposals. This one is titled, “Theory and Applications of Hyper-Heuristics.” What are hyper-heuristics? A comprehensive understanding of the ideas demands some familiarity with meta-heuristic algorithms. So what are meta-heuristic algorithms then? In its simplest manifestation, a meta-heuristic algorithm takes a user-defined problem as an input and starts generating solutions for it. It then improves those solutions iteratively in a meaningful sense, and stops iterating when the desired solution has been found. If you find this explanation bemusing, read it again a couple of times before moving on.
So what is a hyper-heuristic algorithm? A hyper-heuristic algorithm takes a user-specified problem as an input and starts generating algorithms that would solve the problem. The algorithms it generates are in turn expected to generate solutions to the problem in whatever way they deem fit. Just like a meta-heuristic algorithm generates newer solutions to a problem iteratively, a hyper-heuristic algorithm generates newer algorithms iteratively. The only difference is of generation of solutions versus the generation of algorithms respectively. Consequently, a hyper-heuristic algorithm can cook-up newer meta-heuristic on its way to solving the problem. However, it is only the algorithms it creates, actually solve the problem. This is a simple explanation of how a hyper-heuristic works. If you are still bemused about how it works, read through this article again.
Hyper-heuristics have been applied to numerous problems and they have shown promising results. What inspires us most about them is their application to vehicle routing problem in the context of developing fully autonomous, self-coordinating unmanned aerial vehicles. Please read the following research proposal to get an idea about our interests in Theory and Applications of Hyper-Heuristics.
Photo by seier+seier 
Algorithms That Create Algorithms by OptimumT is licensed under a Creative Commons Attribution-NoDerivatives 4.0 International License.