’We learn from models provided by nature’

Information systems specialist Christian Grimme explains the principle of evolutionary algorithms

In many ways, nature serves as a model for processes and functions which we use in our everyday lives. Prof. Christian Grimme from the Department of Information Systems at the University of Mnster has been working for many years now on, and with, so-called evolutionary algorithms which – as the name suggests – are oriented towards the underlying thoughts contained in the theory of biological evolution. Kathrin Kottke spoke to him about the function and the uses of this informatics-based approach.

Most people would place evolution in the field of biology. You are an information systems specialist and you too work with the concept of evolution. How does that fit together’

You’re right – at first glance, these are two very different disciplines, but both sciences – biology and informatics – are interested in the principles of how systems function. And evolution is a fascinating, open, dynamic system. So, in the widest sense, I work with concepts, designs and processes – also known as bionics – which are all’inspired by nature. One well-known example of bionic technologies is dirt-resistant paints, which make use of the lotus effect. Evolutionary algorithms also belong to the category of nature-inspired concepts – except that in these cases nature serves as a model in the development of algorithms.

And what is it about’

It’s a process of optimisation based on the thoughts contained in the modern theory of evolution according to Charles Darwin. However, the concept of evolution is to be understood rather as a metaphor in this connection. The adaptation of animals or plants to their environment functions very well as a result of evolutionary developments and is driven by selection and mutation. In science and in industry we make use of this observation and learn from models in nature. Specifically, evolutionary algorithms are computer applications which imitate biological evolutionary processes using simplified notions of models in order to provide specific solutions to complex problems.

What exactly does that mean’

We first have to remember the principle of the theory of evolution. Put simply, it says that living things are always subject to change and have to adapt in order to survive as a species. In any species, the individuals which have adapted best are at an advantage when they reproduce. These adaptations arise in nature by chance, through constant mutations in the genome. The genome which has adapted best survives as a result of more frequent reproduction. This means that, in the long term, adaptation to external influences leads to new species arising which are better adapted. If we transfer this to optimisation, we could say – again, expressed in a simplified way – that this cycle of adaptation continues until at some point a good solution to a problem has been found, even if it is not perhaps the best one. And that, in principle, describes the general loop which is also used in evolutionary algorithms: repeated mixing and random changes in a population of solutions and selecting the best solutions for a problem.

Can you give examples of applications to make that clearer’

The field of application for evolutionary algorithms is almost unlimited. They are particularly interesting for difficult problems where we don’t know how we can arrive at a good solution – for example, optimising transportation routes for logistics companies or drawing up a machine utilisation plan in a large factory. One of the first industrial applications came from the inventor of evolutionary strategies, the aerospace engineer Hans-Paul Schwefel. He tried to design the optimal interior form of a two-phase jet nozzle with maximum boost. His starting point was a jet form which is funnel-shaped and tapered and which then, again like a funnel, opens out again. He then applied evolutionary algorithms and cut up the jet into small discs, so to speak. The evolutionary algorithm reassembled the discs by changing their order – always with mutation and always selecting the best solutions. Ultimately, this led to a new, surprising form which was significantly better than the initial form.

And what challenges are you tackling with evolutionary algorithms’

In my work I’m investigating so-called multi-objective optimisation. What this means is that that there is not just one objective I’m working towards in the optimisation process, but several. And these objectives often contradict one another. For example: someone wants to buy a car which is particularly safe and which, at the same time, has low fuel consumption. Achieving both at the same time is impossible, as a safe car is often large and heavy and, accordingly, has a higher fuel consumption than small, lightweight cars. These objectives have to be reconciled, or optimum compromises have to be produced.

And how exactly do the algorithms help in this’

Finding the compromises is a difficult problem. Often, mathematics cannot adequately formulate the connections and interactions between the objectives. Therefore, we fall back on evolutionary algorithms. It has been demonstrated that specially adapted methods for these types of problem lead to good solutions. For example, the algorithms calculate optimum compromises in industrial applications or in logistics – going as far as tackling economic and social questions.

This article is from the university newspaper wissen