Evolving the adjacent possible for engineering design

The apparent optimality of evolution in the biological context has inspired the development of the so-called "nature-inspired methods", that can yield near-optimal solutions to difficult problems in engineering. Evolutionary concepts such as selection, mutation, and crossover are employed by Evolutionary Algorithms in (meta)heuristic settings to solve a surprisingly wide gamut of problems in optimisation, design, scheduling, routing, and many other tasks that usually involve the search of very large solution spaces and belong to demanding complexity classes.

As the search of the solution space takes place over the genome level, designs that are similar in the solution space should be encoded such that they are close to one another in the genome space (this is the 'adjacent possible' of the project title), but since the solution space is not known in advance, a co-process to improve the mapping from gene space to solution space has the potential to benefit evolutionary algorithms generally, but particularly in the domain of design. It is also critical to ensure not only efficiency of the search in its scope, but also the lack of information loss while we "enter" or "exit" it.

This means that the encoding of data on a genome, as much as its translation to create solutions in the scope of the problem at hand, are two processes that should optimise.

The aim is to create a dynamic encoding of genotypes that capture optimally the problem space to ultimately produce phenotypes/solutions that meet the design standards in various engineering problems with ease.

This study will focus on analysing the encoding mechanisms observed in biological gene regulatory networks and, more importantly, on drawing inspiration from them to optimise encoding in Evolutionary Algorithms. By representing the problem space efficiently (both in terms of tractability and quality of solutions) it can significantly improve the problem space exploration, without undermining its exploitation.

Alignment to EPSRC's strategies and research areas = Engineering design, ICT Any companies or collaborators involved = Queen's Belfast and Loughborough universities