Peaks

Keynotes

Christine Zarges

Birmingham Fellow & Lecturer
School of Computer Science
University of Birmingham

Tutorial: Fixed Budget Analysis

Abstract

Fixed budget analysis has recently been introduced as an alternative paradigm to runtime analysis in the theory of randomised search heuristics. While in runtime analysis one usually concentrates on the average time needed to find an optimal or approximately optimal solution, in fixed budget analysis we are interested in the quality of the best solution obtained after a pre-specified number of function evaluations called budget. This matches the application of randomised search heuristics in a much better way since usually optimal solutions are not known and even if found cannot be recognised and thus, algorithms are usually stopped after some time. In this tutorial, we review existing work on fixed budget analysis. Starting with a simple and intuitive example we present a case study to demonstrate how fixed budget analysis can give additional insights into the performance of randomised search heuristics. We then give an overview of novel methods tailored to fixed budget analysis, results obtained so far as well as application areas considered in recent publications. We finish by pointing out important aspects of future work.

James Marshall

Professor of Theoretical and Computational Biology
Computational Biology Group
Department of Computer Science
University of Sheffield

Go to the bee and be wise: principled analysis and design of bio-inspired collective decision-making algorithms

Abstract

Engineers frequently draw inspiration from the collective behaviour of social insects to design artificial decision-making algorithms. Such biomimemtic engineering is usually based on the principle that ‘nature knows best’, that millions of years of evolution have optimised observed behaviours, and so engineers may simply copy these to implement effective and efficient artificial systems. Yet the artificial systems designed in this way are often both complicated and complex, and this opacity means that frequently the predominant technique for evaluating their performance is numerical simulation. In this talk I will present the fruits of a different, normative approach, which started with a provably optimal stochastic model of decision-making, that optimise the speed-accuracy trade-off, then investigated how this could be implemented in a distributed decision-making system. I will also describe how observations of a novel signalling behaviour in house-hunting honeybee swarms, coupled with principled dynamical systems and numerical analyses, led to the identification of a highly efficient and effective decision motif, and a reevaluation of the appropriate optimality criterion for natural and artificial decision-making systems. This new criterion should frequently replace the speed-accuracy trade-off, universally invoked by psychologists, neuroscientists, biologists and engineers. Finally I will discuss plans to search for further implementations of this decision motif in other natural systems, from single cells up to primate brains, and to implement these behaviours in highly scaleable swarms of robots.

Jürgen Branke

Professor of Operational Research & Systems
Operational Research & Management Sciences Group
Warwick Business School
University of Warwick

Tracking optima in dynamic environments

Abstract

Many real-world problems change dynamically over time, which makes it necessary to track the changing optimum. Since nature requires continuous adaptation, nature-inspired metaheuristics seem a promising candidate to tackle such problems. However, the standard metaheuristics often perform quite poorly in a dynamic environment, and a lot of new algorithmic variants have been proposed in the past 20 years, specifically targeted at dynamic environments. This talk will provide an overview on the field, its state-of-the-art and challenges, and then focus on some recent research on the use of Efficient Global Optimization for dynamic environments.