Tutorials

Tutorials will be organised at the conference, and will be free to all conference participants.

Note that tutorials (each lasts about 1 hour 40 minutes) will be presented on Monday 15 December 2014.

Confirmed Tutorials


Tutorial 1

Evolving and Designing Neural Network Ensembles Effectively
Professor Xin Yao
University of Birmingham
https://www.cs.bham.ac.uk/~xin/

This tutorial starts with an overview of different evolutionary approaches to learn the weights, architectures and learning rules of neural networks. However, monolithic neural networks become too complex to train and evolve for large and complex problems. It is often better to design a collection of simpler neural networks that work collectively and cooperatively to solve a large and complex problem. The key issue here is how to design such a collection automatically so that it has the best generalisation. This tutorial next describes the motivation of evolving neural network ensembles and explains the potential links between evolving a diverse population of neural networks and designing a neural network ensemble. Negative correlation learning is introduced as an example to illustrate such a link. Inspired by negative correlation learning and evolving ensembles, several improved ensemble learning algorithms, including multi-objective ensemble learning, are also introduced. Some applications examples are given. Finally the tutorial ends with some recent ensemble approaches to online learning, class imbalance learning and semi-supervised learning.

Tutorial 2

How to develop a killer EC-based application?
Emeritus Professor Zbigniew Michalewicz
University of Adelaide
http://www.cs.adelaide.edu.au/~zbyszek

The talk is based on 14 years industry experience — in particular, we will talk about some EC-based applications developed at SolveIT Software that allowed to grow the business from zero to almost 180 employees and $20 million in revenue before selling the business to Schneider Electric. Because of these applications, SolveIT Software became the 3rd fastest-growing company in Australia in 2012, as ranked by Deloitte; the company won numerous awards, and counted among its customers some of the largest corporations in the world, including Rio Tinto, BHP Billiton, and Xstrata.

In this tutorial we will focus on a few features of decision-support software that make the applications "irresistible" ... We will discuss concepts of adaptive business intelligence, dynamic environments, what-if scenarios, trade-off analysis, strategic optimisation, interfaces, and global optimisation in the context of multi-silo problems. The talk will be illustrated by a few pieces of software.

Tutorial 3

Parameterized Complexity Analysis of Bio-Inspired Computing
Associate Professor Frank Neumann
University of Adelaide
http://www.cs.adelaide.edu.au/~frank

In real applications, problem inputs are typically structured or restricted in some way. Evolutionary algorithms and other bio-inspired algorithms can sometimes exploit such extra structure, while in some cases it can be problematic. In any case, from a theoretical perspective, little is understood about how different structural parameters affect the running time of such algorithms.

In this tutorial we present techniques from the new and thriving field of parameterized complexity theory. These techniques allow for a rigorous understanding of the influence of problem structure on the running time of evolutionary algorithms on NP-hard combinatorial optimization problems. We show how these techniques allow one to decompose algorithmic running time as a function of both problem size and additional parameters. In this way, one can attain a more detailed description of what structural aspects contribute to the exponential running time of EAs applied to solving hard problems.

After a general introduction into the computational complexity analysis of bio-inspired computation, we will present detailed and thorough parameterised results for bio-inspired computing on problems such as the traveling salesperson problem and makespan scheduling. We will also outline directions for future research and discuss some open questions.

Presenter Bio: Frank Neumann received his diploma and Ph.D. from the Christian-Albrechts-University of Kiel in 2002 and 2006, respectively. Currently, he is an Associate Professor and leader of the Optimisation and Logistics Group at the School of Computer Science, The University of Adelaide, Australia. Frank is the general chair of ACM GECCO 2016. With Kenneth De Jong he organised ACM FOGA 2013 in Adelaide and together with Carsten Witt he has written the textbook "Bioinspired Computation in Combinatorial Optimization - Algorithms and Their Computational Complexity" published by Springer. He is an Associate Editor of the journal "Evolutionary Computation" (MIT Press), vice-chair of IEEE Task Force on Theoretical Foundations of Bio-inspired Computation, and chair of the IEEE Task Force on Evolutionary Scheduling and Combinatorial Optimization. In his work, he considers algorithmic approaches in particular for combinatorial and multi-objective optimization problems and focuses on theoretical aspects of evolutionary computation as well as high impact applications in the areas of renewable energy, logistics, and sports.

Tutorial 4

Estimation of Distribution Algorithms and Probabilistic Modelling in Evolutionary Computation
Associate Professor Marcus Gallagher
University of Queensland
http://staff.itee.uq.edu.au/marcusg/

Estimation of Distribution Algorithms (EDAs) are a class of evolutionary algorithms that utilize probabilistic modelling and learning techniques to drive the stochastic search process for solving optimization problems. In recent years, EDAs have emerged as a significant class of algorithms, with numerous algorithms proposed for both discrete and continuous spaces. This tutorial will provide an introduction to the fundamental concepts and principles of EDAs, review existing and state of the art algorithms and discuss current directions in the research.

Outline of Material: The main topics to be covered are:

Pseudo code and/or Matlab code will be used and demonstrations of algorithms run during the tutorial.

Presenter Bio: Dr Marcus Gallagher is an Associate Professor in the School of Information Technologyand Electrical Engineering at the University of Queensland, Brisbane Australia. One of his main research interests is in EDAs, where he has published numerous journal and conference papers and has supervised several PhD students. Marcus has 14 years of experience at teaching undergraduate and postgraduate courses at UQ, including a machine learning course. He has given similar presentations in the past, such as an invited talk at the IEEE Summer School of Computational Intelligence and Integrated Technologies (Griffith University, Gold Coast, Australia 2012) and the 2009 Australasian Computational Intelligence Summer School (Melbourne). More information including a list of publications can be found via my homepage: http://staff.itee.uq.edu.au/marcusg/index.html

Tutorial 5

Advances on Evolutionary Many-objective Optimization
Associate Professor Hernán Aguirre
Shinshu University, Japan
http://soar-rd.shinshu-u.ac.jp/profile/en.gNDpbpkh.html

Multi-objective evolutionary algorithms (MOEAs) are widely used in practice for solving multi-objective design and optimization problems. Historically, most applications of MOEAs have dealt with two and three objective problems, leading to the development of several evolutionary approaches that work successfully in these low dimensional objective spaces. Recently, there is a growing interest in industry to solve many-objective optimization problems, where the number of objective functions to optimize simultaneously is more than three. However, conventional MOEAs were not designed to cope with the challenges imposed by many-objective optimization and scale up poorly with the number of objectives of the problem. The development of robust, scalable, many-objective optimizers is an ongoing effort and a promising line of research. Critical to the development of such algorithms is an understanding of fundamental features of many-objective landscapes and the interaction between selection, variation, and population size to appropriately support the evolutionary search in high-dimensional spaces.

This tutorial aims at giving an introduction to evolutionary many-objective optimization, discussing important characteristics of many-objective landscapes and relating them to working principles, performance and behavior of the optimizers. Some of the recent research results will be presented in some detail emphasizing the real world application of many-objective algorithms. More specifically, the tutorial will:

  1. introduce the basic principles of multi-objective evolutionary algorithms,
  2. show scalability issues of conventional multi-objective optimizers when applied to many-objective problems,
  3. introduce important features of many-objective landscapes and show the effectiveness of selection and variation operators when the characteristics of the many-objective landscapes are taken into account,
  4. show the effects of population size,
  5. present a general overview of the approaches to many-objective optimization, together with their state-of-the-art algorithms and techniques,
  6. discuss real world applications of evolutionary many-objective algorithms, and
  7. present open question throughout the tutorial that can serve for all participants as a starting point for future research and/or discussions during the conference.

Expected audience: This tutorial is intended for both novices and regular users of MOEAs. The novice will learn about the foundations of multi- and many-objective optimization and the basic working principles of state-of-the-art MOEAs. Those working on multi- and many-objective optimization will benefit from a unified view between landscape's features and algorithm's working principles. This tutorial will also be of great interest to those working on real world applications of evolutionary many-objective algorithms.

Presenter Bio: Hernán Aguirre received his Engineer degree in computer systems from Escuela Politécnica Nacional, Ecuador, in 1992, and the M.S. and Ph.D. degrees from Shinshu University, Japan, in 2000 and 2003, respectively. Currently, he is an associate professor at Shinshu University. His research interests include evolutionary computation, multidisciplinary design optimization, and sustainability. He has written over 120 international journal and conference research papers on evolutionary algorithms, focusing on the working principles of single-, multi-, and many-objective (any-objective) evolutionary optimizers, landscape analysis, and epistasis. He collaborates actively with industry and with the Japan Aerospace Exploration Agency (JAXA) on the development and application of many-objective evolutionary algorithms to real-world problems.

Tutorial 6

Monte Carlo Tree Search and Evolutionary Enhancements
Professor Simon Lucas
University of Essex
http://dces.essex.ac.uk/staff/lucas/

Monte Carlo tree search (MCTS) is a powerful search method that combines the precision of tree search with the generality of random sampling. It has received considerable interest due to its outstanding success in the challenging board games such as Go and Hex, but has also proved to be a leading method in many other games and some applications beyond games.

In this tutorial I will cover the basics of the algorithm starting with flat Monte Carlo (no tree), then show the benefits of building a tree, and the standard ways of balancing exploration versus exploitation using the Upper Confidence Bounds for Trees (UCT) formula. Despite the theoretical appeal of UCT, many MCTS programs rely more heavily on heuristics, so examples of these are also included. I'll then explore some cases such as real-time games and control problems where standard MCTS can perform poorly, and show ways in which evolution can be used to tune the algorithm to achieve good performance.

The tutorial will include many demonstrations to help explain the key points, and snippets of code / pseudocode will be explained to provide a practical understanding of the algorithm.

A complete implementation in a high level language will also be provided so that delegates can take away some working programs ready to apply to their own problems.

Audience: The tutorial is aimed at students and researchers with a solid knowledge of computational intelligence but little knowledge of MCTS.

Presenter Bio: Simon Lucas is a full professor of computer science at the University of Essex where he leads the Game Intelligence Group. His main research interests are games, evolutionary algorithms and machine learning and he has published widely in these fields with more than 170 peer reviewed publications. He co-founded the IEEE Conference on Computational Intelligence and Games and is the founding editor-in-chief of the IEEE Transactions on Computational Intelligence and AI in Games. He is the Essex PI for the IGGI Centre for Doctoral Training: http://www.iggi.org.uk/

He recently developed Griddle, a fun word game available (in early access form) on Google Play:
https://play.google.com/store/apps/details?id=com.lucapps.state