Tutorials

A Gentle Introduction to Theory (for Non-Theoreticians)

This tutorial addresses GECCO attendees who do not regularly use
theoretical methods in their research. For these, we give a smooth
introduction to the theory of evolutionary computation.
Complementing other introductory theory tutorials, we do not discuss
mathematical methods or particular results, but explain

- what theory of evolutionary algorithms aims at,
- how research in theory of evolutionary computation is done,
- how to interpret statements from the theory literature,
- what are some important contributions of theory to our field,
- and what are the current hot topics.

Evolution of Neural Networks

Evolution of artificial neural networks has recently emerged as a powerful technique in two areas. First, while the standard value-function based reinforcement learning works well when the environment is fully observable, neuroevolution makes it possible to disambiguate hidden state through memory. Such memory makes new applications possible in areas such as robotic control, game playing, and artificial life. Second, deep learning performance depends crucially on the network architecture and hyperparameters. While many such architectures are too complex to be optimized by hand, neuroevolution can be used to do so automatically. Such evolutionary AutoML can be used to achieve good deep learning performance even with limited resources, or state=of-the art performance with more effort. It is also possible to optimize other aspects of the architecture, like its size, speed, or fit with hardware. In this tutorial, I will review (1) neuroevolution methods that evolve fixed-topology networks, network topologies, and network construction processes, (2) methods for neural architecture search and evolutionary AutoML, and (3) applications of these techniques in control, robotics, artificial life, games, image processing, and language.

NEW Evolutionary Computation and Games

In recent years, research in AI for Games and Games for AI has enjoyed
rapid progress and a sharp rise in popularity. In this field, various
kinds of AI algorithms are tested on benchmarks based on e.g. board
games and video games, and new AI-based solutions are developed for
problems in game development and game design. This tutorial will give
an overview of key research challenges and methods of choice in
evolutionary computation applied to games. The tutorial is divided into
two parts, where the second part builds on methods and results
introduced in the first part. The first part will focus on
evolutionary computation methods for playing games, including
neuroevolution and evolutionary planning. The second part will focus
on the use of evolutionary computation for game testing and procedural
content generation, as well as player experience prediction and game adaptation.

Evolutionary Computation: A Unified Approach

The field of Evolutionary Computation has experienced tremendous growth over the past 20 years, resulting in a wide variety of evolutionary algorithms and applications. The result poses an interesting dilemma for many practitioners in the sense that, with such a wide variety of algorithms and approaches, it is often hard to see the relationships between them, assess strengths and weaknesses, and make good choices for new application areas.
This tutorial is intended to give an overview of a general EC framework that can help compare and contrast approaches, encourages crossbreeding, and facilitates intelligent design choices. The use of this framework is then illustrated by showing how traditional EAs can be compared and contrasted with it, and how new EAs can be effectively designed using it.
Finally, the framework is used to identify some important open issues that need further research.

Evolutionary Many-Objective Optimization

The goal of the tutorial is clearly explaining difficulties of evolutionary many-objective optimization, approaches to the handling of those difficulties, and promising future research directions. Evolutionary multi-objective optimization (EMO) has been a very active research area in the field of evolutionary computation in the last two decades. In the EMO area, the hottest research topic is evolutionary many-objective optimization. The difference between multi-objective and many-objective optimization is simply the number of objectives. Multi-objective problems with four or more objectives are usually referred to as many-objective problems. It sounds that there exists no significant difference between three-objective and four-objective problems. However, the increase in the number of objectives significantly makes multi-objective problems difficult. In the first part (Part I: Difficulties), we clearly explain not only frequently-discussed well-known difficulties such as the weakening selection pressure towards the Pareto front and the exponential increase in the number of solutions for approximating the entire Pareto front but also other hidden difficulties such as the deterioration of the usefulness of crossover and the difficulty of performance evaluation of solution sets. The attendees of the tutorial will learn why many-objective optimization is difficult for EMO algorithms. After the clear explanations about the difficulties of many-objective optimization, we explain in the second part (Part II: Approaches and Future Directions) how to handle each difficulty. For example, we explain how to prevent the Pareto dominance relation from weakening its selection pressure and how to prevent a binary crossover operator from decreasing its search efficiently. We also explain some state-of-the-art many-objective algorithms. Some promising research directions are explained in detail in the tutorial.

Evolutionary Multi-objective Optimization: Past, Present and Future

Genetic algorithms (GAs) were first formally demonstrated to solve optimization problems head-to-head with classical point-based methods in 1975 by Kenneth De Jong. Goldberg and Richardson demonstrated the use of GAs for finding multiple optimal solutions for a single-objective multi-modal function in 1987. All these early studies led to the suggestion of three evolutionary multi-objective optimization (EMO) algorithms during 1993-95, following a suggestion by Goldberg, which marked the beginning of the EMO field. Now, there are more than 7,000 papers and reports generated every year in EMO with more than 50% of them being applied to outside computer science and engineering area. About 25% of IEEE TEVC published papers are in EMO area. The most cited paper in the whole GEC field, having more than 30,000 Google Scholar citations, comes from EMO field and three of top five most popular papers of IEEE TEVC and four our of five most cited papers from MIT Press's Evol. Computation Journal are from EMO area. There are at least three major software companies which survive on EMO algorithms. Every day, EMO is attracting new researchers and practitioners into the field.
In this tutorial, we shall provide a systematic and chronological account of how EMO field was started, details of a few key EMO algorithms that made the field popular, and key applications which show-case their practical importance. We shall discuss in details key current research topics which will motivate new-comers to get started with directions. Some of the topics which would be covered are: evolutionary many-objective optimization, surrogate-assisted EMO, robust and reliability based EMO, EMO with decision-making, Multiobjectivization, EMO based knowledge extraction, theoretical advancements of EMO, and others. Finally, we shall discuss the presenter's account on immediate and futuristic research ideas of the field. Some of these topics will include EMO for dynamic problems, EMO for bilevel problems, EMO for machine learning including CNN and DNN architecture search and EMO for very large-scale problems. The tutorial will be concluded by pointing to number of different resources (books, public domain codes, key websites, and others) and by demonstrating working of a few public domain codes.

Hyper-heuristics

Hyper-heuristics
The automatic design of algorithms has been an early aim of both machine learning and AI, but has proved elusive. The aim of this tutorial is to introduce hyper-heuristics as a principled approach to the automatic design of algorithms. Hyper-heuristics are metaheuristics applied to a space of algorithms; i.e., any general heuristic method of sampling a set of candidate algorithms. In particular, this tutorial will demonstrate how to mine existing algorithms to obtain algorithmic primitives for the hyper-heuristic to compose new algorithmic solutions from, and to employ various types of genetic programming to execute the composition process; i.e., the search of program space.

This tutorial will place hyper-heuristics in the context of genetic programming - which differs in that it constructs solutions from scratch using atomic primitives - as well as genetic improvement - which takes a program as starting point and improves on it (a recent direction introduced by William Langdon).

The approach proceeds from the observation that it is possible to define an invariant framework for the core of any class of algorithms (often by examining existing human-written algorithms for inspiration). The variant components of the algorithm can then be generated by genetic programming. Each instance of the framework therefore defines a family of algorithms. While this allows searches in constrained search spaces based on problem knowledge, it does not in any way limit the generality of this approach, as the template can be chosen to be any executable program and the primitive set can be selected to be Turing-complete. Typically, however, the initial algorithmic primitive set is composed of primitive components of existing high-performing algorithms for the problems being targeted; this more targeted approach very significantly reduces the initial search space, resulting in a practical approach rather than a mere theoretical curiosity. Iterative refining of the primitives allows for gradual and directed enlarging of the search space until convergence.

This leads to a technique for mass-producing algorithms that can be customised to the context of end-use. This is perhaps best illustrated as follows: typically a researcher might create a travelling salesperson algorithm (TSP) by hand. When executed, this algorithm returns a solution to a specific instance of the TSP. We will describe a method that generates TSP algorithms that are tuned to representative instances of interest to the end-user. This method has been applied to a growing number of domains including; data mining/machine learning, combinatorial problems including bin packing (on- and off-line), Boolean satisfiability, job shop scheduling, exam timetabling, image recognition, black-box function optimization, wind-farm layout, and the automated design of meta-heuristics themselves (from selection and mutation operators to the overall meta-heuristic architecture).

This tutorial will provide a step-by-step guide which takes the novice through the distinct stages of automatic design. Examples will illustrate and reinforce the issues of practical application. This technique has repeatedly produced results which outperform their manually designed counterparts, and a theoretical underpinning will be given to demonstrate why this is the case. Automatic design will become an increasingly attractive proposition as the cost of human design will only increase in-line with inflation, while the speed of processors increases in-line with Moore's law, thus making automatic design attractive for industrial application. Basic knowledge of genetic programming will be assumed.

Introduction to Genetic Programming

Genetic programming emerged in the early 1990's as one of the most exciting new evolutionary algorithm paradigms. It has rapidly grown into a thriving area of research and application. While sharing the evolutionary inspired algorithm principles of a genetic algorithm, it differs by exploiting an executable genome. Genetic programming evolves a 'program' to solve a problem rather than a single solution. This tutorial introduces the basic genetic programming paradigm. It explains how the powerful capability of genetic programming is derived from modular algorithmic components: executable representations such as a parse tree, variation operators that preserve syntax and explore a variable length, hierarchical solution space, appropriately chosen programming functions and fitness function specification. It provides demos and walks through an example of GP software.

Introductory Mathematical Programming for EC

Global optimization of complex models has been for several decades approached by means of formal algorithms as well as Mathematical Programming (MP) (often branded as Operations Research, yet strongly rooted at Theoretical CS), and simultaneously has been treated by a wide range of dedicated heuristics (frequently under the label of Soft Computing) – where EC resides. The former is applicable when explicit modeling is available, whereas the latter is typically utilized for simulation- or experimentation-based optimization (but also applicable for explicit models). These two branches complement each other, yet practically studied under two independent CS disciplines.
It is widely recognized nowadays that EC scholars become stronger, better-equipped researchers when obtaining knowledge on this so-called "optimization complement". In other words, the claim that our EC education should encompass basic MP is untenable at present times, and this tutorial aims at bridging the gap for our community's scholars.
The tutorial comprises three parts, aiming to introduce basic MP for EC audience.
The first part introduces the fundamentals of MP. It overviews mathematical optimization and outlines its taxonomy when classified by the underlying model: Convex Optimization (linear programs (pure-LP) and non-linear programs), versus Combinatorial Optimization (integer and mixed-integer linear programs (M)ILP, integer quadratic programs IQP). It then discusses some of the theoretical aspects, such as polyhedra and the duality theorem.
The second part focuses on MP in practice. The tutorial presents the principles of MP modeling, with emphasis on the roles of constraints and auxiliary/dummy decision variables in MP. It is compared to equivalent modeling for EC heuristics, which operate differently with respect to such components. It then covers selected algorithms for the major classes of problems (Dantzig's Simplex for pure-LP, Ellipsoid for convex models, and Branch-and-bound for ILP).
The third part constitutes an interactive demo session of problem-solving using IBM's CPLEX.

Learning Classifier Systems: From Principles to Modern Systems

Learning Classifier Systems (LCSs) emerged in the late 1970s and since then have attracted a lot of research attention. Originally introduced as a technique to model adaptive agents within Holland’s notion of complex adaptive systems, various enhancements toward a full-fledged Machine Learning (ML) technique with an evolutionary component at its core have appeared. Nowadays, several variations capable of dealing with most modern ML tasks exist, including online and batch-wise supervised learning, single- and multi-step reinforcement learning, and even unsupervised learning. This great flexibility, which is due to their modular and clearly defined algorithmic structure paving the way for simple and straight-forward adaptations, is unique in the field of Evolutionary Machine Learning (EML) – yielding the LCS paradigm an indisputable permanent place in the EML field. Despite this well-known blueprint comprising the building blocks bringing LCSs to function, gaining theoretical insights regarding the interplay between them has been a crucial research topic for a long time, and this still constitutes a pursued subject of active research.

In this tutorial, the main goal is to introduce exactly these building blocks of LCS-based EML and to conceptually develop a modern, generic Michigan-style LCS step by step from scratch. The resulting generic LCS is then brought in line with the mostly investigated representative – Wilson’s XCS Classifier System (or simply XCS) – which serves as reference system for the subsequent parts.
In order to provide a holistic view on LCSs, the tutorial also provides a sketch on the lineage and historical developments in the first part, but will then quickly focus on the more prominent Michigan-style systems.

NEW: Recent theoretical advances in XCS research will be the subject of the tutorial’s second part. The attendees should get in touch with the fundamental challenges and also the bounds of what can be achieved by XCS under which circumstances.

The third part of the tutorial is then devoted to the state of the art in LCS research in terms of their real world applicability. The most recent advances that have led to modern systems such as XCSF for online function approximation or ExSTraCS for large-scale supervised data mining will thus be the subject of discussion.
The tutorial closes with a wrap-up regarding the distinctive potentials of LCS and with suggestions for a complementary (“traditional”) ML-centric view on this unique paradigm. With the intention to provide the audience with an impression about where LCS research stands these days and which open questions are still around, a brief review of most recent endeavors to tackle unsolved issues closes the tutorial.

Model-based Evolutionary Algorithms

In model-based evolutionary algorithms the variation operators are guided by the
use of a model that conveys problem-specific information so as to increase the
chances that combining the currently available solutions leads to improved
solutions. Such models can be constructed beforehand for a specific problem, or
they can be learnt during the optimization process. Well-known types of
algorithms of the latter type are Estimation-of-Distribution Algorithms (EDAs)
where probabilistic models of promising solutions are built. Samples are
subsequently drawn from these models to generate new solutions.

In general, replacing traditional crossover and mutation operators by building
and using models enables the use of machine learning techniques for automatic
discovery of problem regularities and subsequent exploitation of these
regularities, thereby enabling the design of optimization techniques that can
automatically adapt to a given problem. This is an especially useful feature
when considering optimization in a black-box setting. The use of models can
furthermore also have major implications for grey-box settings where not
everything about the problem is considered to be unknown a priori.

Successful applications include Ising spin glasses in 2D and 3D, graph
partitioning, MAXSAT, scheduling, feature construction, windfarm layouting, and
cancer radiation treatment optimization.

Although EDAs are the best known example of model-based EAs, other, including
more recent, approaches exist. Of particular interest is the family of Optimal
Mixing EAs (which includes the Linkage Tree Genetic Algorithm and various other
GOMEA variants). The tutorial will mainly focus on these types of MBEAs.

Neuroevolution for Deep Reinforcement Learning Problems

In recent years, there has been a resurgence of interest in reinforcement learning (RL), particularly in the deep learning community. While much of the attention has been focused on using Value-function learning approaches (i.e. Q-Learning) or Estimated Policy Gradient-based approaches to train neural-network policies, little attention has been paid to Neuroevolution (NE) for policy search. The larger research community may have forgotten about previous successes of Neuroevolution 1.

Some of the most challenging reinforcement learning problems are those where reward signals are sparse and noisy. For many of these problems, we only know the outcome at the end of the task, such as whether the agent wins or loses, whether the robot arm picks up the object or not, or whether the agent has survived. Since NE only require the final cumulative reward that an agent gets at the end of its rollout in an environment, these are the types of problems where NE may have an advantage over traditional RL methods.

In this tutorial, I show how Neuroevolution can be successfully applied to Deep RL problems to help find a suitable set of model parameters for a neural network agent. Using popular modern software frameworks for RL (TensorFlow, OpenAI Gym, pybullet, roboschool), I will apply NE to continuous control robotic tasks, and show we can obtain very good results to control bipedal robot walkers, Kuka robot arm for grasping tasks, Minitaur robot, and also various existing baseline locomotion tasks common in the Deep RL literature. I will even show that NE can even obtain state-of-the-art results 2 over Deep RL methods, and highlight ways to use NE that can lead to more stable and robust policies compared to traditional RL methods. I will also describe how to incorporate NE techniques into existing RL research pipelines taking advantage of distributed processing on Cloud Compute 2, 3.

I will also discuss how to combine techniques from deep learning, such as the use of deep generative models, with Neuroevolution to solve more challenging Deep Reinforcement Learning problems that rely on high dimensional video inputs for continous robotics control, or for video game simulation tasks. We will look at combining model-based reinforcement learning approaches with Neuroevolution to tackle these problems, using TensorFlow, OpenAI Gym, and pybullet environments.

Lastly, we will cover recent developments where we have seen a lot of success in the past 2 years in areas where Neuroevolution has incorporated concepts from Deep Learning/RL, and vice versa. I will discuss recent topics that explore the use of Neural Network Topology evolution to find architectures that have strong inductive biases for RL tasks, and show that such architectures can still work without learning the weight parameters of such a network.

1 Risto Miikkulainen’s Slides on Neuroevolution.

http://nn.cs.utexas.edu/downloads/slides/miikkulainen.ijcnn13.pdf

2 Evolving Stable Strategies

http://blog.otoro.net/2017/11/12/evolving-stable-strategies/

3 Evolution Strategies as a Scalable Alternative to Reinforcement Learning

https://arxiv.org/abs/1703.03864

Representations for Evolutionary Algorithms

Successful and efficient use of evolutionary algorithms (EA) depends on the choice of the genotype, the problem representation (mapping from genotype to phenotype) and on the choice of search operators that are applied to the genotypes. These choices cannot be made independently of each other. The question whether a certain representation leads to better performing EAs than an alternative representation can only be answered when the operators applied are taken into consideration. The reverse is also true: deciding between alternative operators is only meaningful for a given representation.

Research in the last few years has identified a number of key concepts to analyse the influence of representation-operator combinations on EA performance. Relevant concepts are the locality and redundancy of representations.

Locality is a result of the interplay between the search operator and the genotype-phenotype mapping. Representations have high locality if the application of variation operators results in new solutions similar to the original ones. Representations are redundant if the number of phenotypes exceeds the number of possible genotypes. Redundant representations can lead to biased encodings if some phenotypes are on average represented by a larger number of genotypes or search operators favor some kind of phenotypes.

The tutorial gives a brief overview about existing guidelines for representation design, illustrates the different aspects of representations, gives a brief overview of models describing the different aspects, and illustrates the relevance of the aspects with practical examples.

It is expected that the participants have a basic understanding of EA principles.

Runtime Analysis of Population-based Evolutionary Algorithms

Populations are at the heart of evolutionary algorithms (EAs). They provide the genetic variation which selection acts upon. A complete picture of EAs can only be obtained if we understand their population dynamics. A rich theory on runtime analysis (also called time-complexity analysis) of EAs has been developed over the last 20 years. The goal of this theory is to show, via rigorous mathematical means, how the performance of EAs depends on their parameter settings and the characteristics of the underlying fitness landscapes.
Initially, runtime analysis of EAs was mostly restricted to simplified EAs that do not employ large populations, such as the (1+1) EA. This tutorial introduces more recent techniques that enable runtime analysis of EAs with realistic population sizes.

The tutorial begins with a brief overview of the population-based EAs that are covered by the techniques. We recall the common stochastic selection mechanisms and how to measure the selection pressure they induce. The main part of the tutorial covers in detail widely applicable techniques tailored to the analysis of populations. We discuss random family trees and branching processes, drift and concentration of measure in populations, and level-based analyses.

To illustrate how these techniques can be applied, we consider several fundamental questions: When are populations necessary for efficient optimisation with EAs? What is the appropriate balance between exploration and exploitation and how does this depend on relationships between mutation and selection rates? What determines an EA's tolerance for uncertainty, e.g. in form of noisy or partially available fitness?

NEW Theoretical Foundations of Evolutionary Computation for Beginners and Veterans (cancelled)

While theory in the last 10 years has largely focused on runtime analysis for polynomial time problems, there exists more than 40 years of theoretical research in evolutionary computation, most of which has nothing to do with runtime analysis. For example, it is not widely known that the behavior of an Evolutionary Algorithm can be influenced by attractors that exists outside the space of the feasible population. The tutorial will mainly focus on the application of evolutionary algorithms to combinatorial problems.

This talk will cover some of the classic theoretical results from the field of evolutionary algorithms, as well as more general theory from operations research and mathematical methods for optimization. The tutorial will review pseudo-Boolean optimization, as well as the representation of functions as both multi-linear polynomials and Fourier polynomials. It will also explain how every pseudo-Boolean optimization problem can be converted into a k-bounded form. And for every k-bounded pseudo-Boolean optimization problem, the location of improving moves (i.e., bit flips) can be computed in constant time, making simple mutation operators unnecessary.

The tutorial will cover 1) No Free Lunch (NFL), 2) Sharpened No Free Lunch and 3) Focused No Free Lunch, and how the different theoretical proofs can lead to seemingly different and even contradictory conclusions. (What many researchers know about NFL might actually be wrong.)
The tutorial will also cover the relationship between functions and representations, the space of all possible representations and the duality between search algorithm and landscapes. This perspective is critical to understanding landscape analysis, landscape visualization, variable neighborhood search methods, memetic algorithms, and self-adaptive search methods.

Other topics include both infinite and finite models of population trajectories. The tutorial will explain both Elementary Landscapes and eigenvectors of search neighborhoods in simple terms and explain how the two are related. Example domains include classic NP-Hard problems such as Graph Coloring and MAX-kSAT.

Finally, the tutorial will also offer a cautionary critique of theory. While theoretical results are typically based on proofs, virtually all proofs are based on assumptions. And yet, theory is sometimes leveraged far beyond the supporting assumptions, which can lead to misleading claims and false conclusions. Multiple examples have occurred over and over again in the field of Evolutionary Computation. Every researcher in the field of Evolutionary Computation needs to be a wiser consumer of both theoretical and empirical results.

NEW A Hands-on Guide to Distributed Computing Paradigms for Evolutionary Computation

Recent advances in machine learning are consistently enabled by increasing amounts of computation. When such computation is beyond the capacity of a single computer, people rely on distributed computing that scales large, parallelizable computing loads across multiple computers for better efficiency and performance. Evolutionary algorithms (EAs) often consist of components and submodules that can be executed in a parallel or asynchronous manner, and therefore, could benefit tremendously from following suitable distributed computing paradigms. However, architecting and setting up parallel algorithms to run over multiple computers are often non-trivial, which adds difficulty for researchers to efficiently utilize computational resources available over clusters and/or cloud.

This tutorial aims to address this gap between efficient algorithm development and scalable distributed execution of algorithms. It provides a hands-on guide on distributed computing paradigms that are suitable for EAs and reinforcement learning (RL) algorithms, and introduces tools and libraries that allow researchers to easily switch between developing algorithms on a local machine and launching the execution of their algorithms in a distributed fashion across multiple machines for best performance.

The tutorial begins with a review of common computational patterns in typical EAs and RL algorithms, and then highlights challenges that these algorithms pose to distributed computing frameworks in terms of reliability, efficiency, and user-friendliness. Next, we briefly review and compare a few classic general-purpose distributed computing frameworks (e.g., MPI, Ipyparallel, etc.) and explain their limitations when applied to EAs and RL. We then dive into more recent frameworks (such as Ray and Fiber, etc.), explain how they overcome the limitations of classic ones, review their design and usage, and compare their performance on benchmark examples.

The second part of the tutorial is a live demo/interactive session of Fiber, a Python library for distributed computing developed and used at Uber AI to support research in evolutionary computation and RL. We will demonstrate how to install and set up Fiber, how to architect and adapt Python programs that run on a single machine to run in a distributed manner over multiple machines with Fiber's APIs, how to visualize the progresses and monitor the health of distributed jobs. We will also demonstrate advanced use cases on how Fiber can be used along with an evolutionary computation framework (DEAP) and typical deep learning frameworks (PyTorch, TensorFlow, and JAX), and show examples of scaling the execution of some classic EAs (such as evolution strategies, genetic algorithms), RL algorithms, and other use cases, e.g., Population-Based Training.

NEW Benchmarking and Analyzing Iterative Optimization Heuristics with IOHprofiler

IOHprofiler is a new benchmarking environment that has been developed for a highly versatile analysis of iterative optimization heuristics (IOHs) such as evolutionary algorithms, local search algorithms, model-based heuristics, etc. A key design principle of IOHprofiler is its highly modular setup, which makes it easy for its users to add algorithms, problems, and performance criteria of their choice. IOHprofiler is also useful for the in-depth analysis of the evolution of adaptive parameters, which can be plotted against fixed-targets or fixed-budgets. The analysis of robustness is also supported.

IOHprofiler supports all types of optimization problems, and is not restricted to a particular search domain. A web-based interface of its analysis procedure is available athttp://iohprofiler.liacs.nl/, the tool itself is available onGitHub (https://github.com/IOHprofiler/IOHanalyzer) and as CRAN package (https://cran.rstudio.com/web/packages/IOHanalyzer/index.html).

The tutorial addresses all GECCO participants interested in analyzing and comparing heuristic solvers. By the end of the tutorial, the participants will known how to benchmark different solvers with IOHprofiler, which performance statistics it supports, and how to contribute to its design.

Decomposition Multi-Objective Optimisation: Current Developments and Future Opportunities

Evolutionary multi-objective optimisation (EMO) has been a major research topic in the field of evolutionary computation for many years. It has been generally accepted that combination of evolutionary algorithms and traditional optimisation methods should be a next generation multi-objective optimisation solver. As the name suggests, the basic idea of the decomposition-based technique is to transform the original complex problem into simplified subproblem(s) so as to facilitate the optimisation. Decomposition methods have been well used and studied in traditional multi-objective optimisation. MOEA/D decomposes a multi-objective problem into a number of subtasks, and then solves them in a collaborative manner. MOEA/D provides a very natural bridge between multi-objective evolutionary algorithms and traditional decomposition methods. It has been a commonly used evolutionary algorithmic framework in recent years.

Within this tutorial, a comprehensive introduction to MOEA/D will be given and selected research results will be presented in more detail. More specifically, we are going to (i) introduce the basic principles of MOEA/D in comparison with other two state-of-the-art EMO frameworks, i.e., Pareto- and indicator-based frameworks; (ii) present a general overview of state-of-the-art MOEA/D variants and their applications; (iii) discuss the future opportunities for possible further developments.

The intended audience of this tutorial can be both novices and people familiar with EMO or MOEA/D. In particular, it is self-contained that foundations of multi-objective optimisation and the basic working principles of EMO algorithms will be included for those without experience in EMO to learn. Open questions will be posed and highlighted for discussion at the latter session of this tutorial.

NEW Design Principles for Matrix Adaptation Evolution Strategies

Covariance Matrix Adaptation Evolution Strategies are regarded as
state-of-the-art in evolutionary real-valued parameter optimization.
These strategies learn the covariance matrix in order to generate
suitable mutations that allow for an efficient approximation of the
optimizer of the problem to be considered. Only recently it has been
shown that there is no need to learn the covariance matrix. Instead
the mutation matrix can be learned directly. That is, one can remove
the covariance matrix stuff from the Evolution Strategy (ES)
without performance degradation. As a result, the algorithms get
simpler and can be easily modified to also tackle large scale
optimization problems and constrained optimization problems.
One such ES was the winner in the 2018 CEC constrained optimization
benchmark competition regarding the high-dimensional problem instances.

This tutorial provides a gentle introduction into Matrix Adaptation
Evolution Strategies (MA-ES) explaining the design principles.
Being based on these principles, it will be shown how one can
modify the MA-ES in order to
a) incorporate self-adaptive behavior,
b) handle large scale optimization problems with thousands of variables
c) treat constrained optimization problems

The tutorial will also show some 2D and 3D graphical demonstrations of
the working of MA-ES handling restricted high-dimensional optimization
problems with non-linear equality constraints.

Dynamic Control Parameter Choices in Evolutionary Computation

One of the most challenging problems in solving optimization problems with evolutionary algorithms is the selection of the control parameters, which allow to adjust the behaviour of the algorithms to the problem at hand. Several control parameters need to be set, for the procedure of searching for the optimum of an objective function to be successful. Suitable control parameter values need to be found, for example, for the population size, the mutation strength, the crossover rate, the selective pressure, etc. The choice of these parameters can have a significant impact on the performance of the algorithm and need thus to be executed with care.

In the early years of evolutionary computation there had been a quest to determine universally "optimal" control parameter choices. At the same time, researchers have soon realized that different parameter settings can be optimal at different stages of the optimization process: at the beginning of an optimization process, one may want to allow a larger mutation rate to increase the chance of finding the most promising regions of the search space ("exploration" phase), while later on, a small mutation rate guarantees the search to stay focused within the promising area ("exploitation" phase). Such dynamic parameter choices are today standard in continuous optimization. However, the situation is much different in discrete optimization, where non-static parameter choices have never lived up to their impact, yet.

The ambition of this tutorial is to contribute to a paradigm change towards a more systematic use of dynamic parameter choices. To this end, we survey existing techniques to automatically select control parameter values on the fly. We will discuss both theoretical and experimental results that demonstrate the unexploited potential of non-static parameter choices. Our tutorial thereby addresses experimentally as well as theory-oriented researchers alike.

Evolutionary Computation for Digital Art

Evolutionary algorithms have been used in various ways to create or guide the creation of digital art. Artificial intelligence is substantially changing the nature of creative processes.
In this tutorial we present techniques from the thriving field of biologically inspired art. We show how evolutionary computation methods and neural networks can be used to enhance artistic creativity and lead to software systems that help users to create artistic work.

We start by providing a general introduction into the use of evolutionary computation methods for digital art and highlight different application areas. This covers different evolutionary algorithms including genetic programming for the creation of artistic images.
Afterwards, we discuss evolutionary algorithms to create artistic artwork in the context of image transition and animation. We show how the connection between evolutionary computation methods and a professional artistic approach finds application in digital animation and new media art, and discuss the different steps of involving evolutionary algorithms for image transition into the creation of paintings. Furthermore, we show how neural networks can be utilised as an inspiration for creating original images.

Subsequently, we give an overview on the use of aesthetic features to evaluate digital art. The feature-based approach complements the existing evaluation through human judgments/analysis and allows to judge digital art in a quantitative way. Finally, we outline directions for future research and discuss some open problems.

NEW Fitness landscape analysis to understand and predict algorithm performance for single- and multi-objective optimization

Many evolutionary and general-purpose search algorithms have been proposed for solving a broad range of single- and multi-objective optimization problems. Despite their skillful design, it is still difficult to achieve a high-level fundamental understanding of why and when an algorithm can be expected to be successful. From an engineering point of view, setting up a systematic and principled methodology to select and/or design an effective search algorithm is also a challenging issue which is attracting more and more attention from the research community. In this context, fitness landscape analysis is a well-established field for understanding the relation between the structure underlying a given problem search space, and the algorithm(s), and the underlying components and/or parameters, being considered for tackled the problem.
Starting by presenting state-of-the-art tools from single-objective fitness landscapes, we identify their main differences and then new additional properties to be addressed for a deep understanding of multi-objective fitness landscapes.
We expose and contrast the impact of fitness landscape geometries on the performance of optimization algorithms for single- and multi-objective optimization problems. A sound and concise summary of features characterizing the structure of a problem instance are identified in particular for multi-objective optimization. We also review the fundamental principles which allows to design new relevant features, and we show the main methodologies to sample combinatorial, and continuous search spaces.
By providing effective tools and practical examples for both single- and multi-objective fitness landscape analysis, further insights are given on the importance of ruggedness, multimodality, and objectives correlation to predict the performances of an instance for optimization problems and algorithms.
At last, we conclude with guidelines for the design of randomized search heuristics based on the main fitness landscape features, and we identify a number of open challenges for the future of fitness landscapes and evolutionary algorithms.

NEW Genetic improvement: Taking real-world source code and improving it using genetic programming.

Genetic Programming (GP) has been on the scene for around 25 years. Genetic Improvement (GI) 1 is “the new kid on the block”. What does GI have to offer over GP? The operational difference is that GI deals with source code, rather than a simulation of code. In other words, GI operates directly on Java or C code, for example, whereas GP typically operates on some tiny subset of a programming language, defined by the function set and terminal set. Another fundamental difference is that GI starts with real-world software (which is in use), whereas GP typically tries to evolve programs from scratch (which is not in use).

These differences may not seem important, as we can still generate the same set of functions; however this subtle difference opens up a vast number of new possibilities for research and this will make GI attractive for industrial applications. Furthermore we can optimize the physical properties of code such as power consumption, size of code, bandwidth, and other non-functional properties, including execution time.

The aim of the tutorial is to

• examine the motives for evolving source code directly, rather than a language built from a function set and terminal set which has to be interpreted after a program has been evolved
• understand different approaches to implementing genetic improvement including operating directly on text files, and operating on abstract syntax trees
• appreciate the new research questions that can be addressed while operating on actual source code
• understand some of the issues regarding measuring non-functional properties such as execution time and power consumption
• examine some of the early examples of genetic improvement and our flagship application will be the world’s first implementation of GI in a live system (this technique has found and fixed all 40 bugs in its first 6 months while operating in a medical facility)
• understanding links between GI and other techniques such as hyper-heuristics, automatic parameter tuning, and deep parameter tuning
• highlight some of the multi-objective research where programs have been evolved that lie on the Pareto front with axes representing different non-functional properties
• give an introduction to GI in No Time - an open source simple micro-framework for GI (https://github.com/gintool/gin).

NEW Quality-Diversity Optimization

A fascinating aspect of natural evolution is its ability to produce a diversity of organisms that are all high performing in their niche. By contrast, the main artificial evolution algorithms are focused on pure optimization, that is, finding a single high-performing solution. 


Quality-Diversity optimization (or illumination) is a recent type of evolutionary algorithm that bridges this gap by generating large collections of diverse solutions that are all high-performing. This concept was introduced by the ``Generative and Developmental Systems community between 2011 (Lehman & Stanley, 2011) and 2015 (Mouret and Clune, 2015) with the ``Novelty Search with Local Competition and ``MAP-Elites'' evolutionary algorithms. The main differences with multi-modal optimization algorithms are that (1) Quality Diversity typically works in the behavioral space (or feature space), and not in the genotypic space, and (2) Quality Diversity attempts to fill the whole behavior space, even if the niche is not a peak in the fitness landscape. In the last 5 years, more than 75 papers have been written about quality diversity, many of them in the GECCO community (a non-exhaustive list is available at https://quality-diversity.github.io/papers).


The collections of solutions obtained by Quality Diversity algorithms open many new applications for evolutionary computation. In robotics, it was used to create repertoires of behaviors (Cully & Mouret, 2016), to allow robots to adapt to damage in a few minutes (Cully & et al. 2015, Nature); in engineering, it can be used to propose a diversity of optimized aerodynamic shapes (Gaier et al., 2018 --- best paper of the CS track); they were also recently used in video games (Khalifa et al., 2018) and for Workforce Scheduling and Routing Problems (WSRP) (Urquhart & Hart, 2018).




This tutorial will give an overview of the various questions addressed in Quality-Diversity optimization, relying on examples from the literature. Past achievements and major contributions, as well as specific challenges in Quality-Diversity optimization, will be described. The tutorial will in particular focus on:



• what is Quality-Diversity optimization?
• similarities and differences with traditional evolutionary algorithms;
• existing variants of Quality-Diversity algorithms;
• example of application: Learning behavioral repertoires in robotics, evolving 3D shapes for aerodynamic design;
• open questions and future challenges.





The tutorial will effectively complement the Complex Systems track, which usually contains several papers about Quality-Diversity algorithms. For instance, 36% (4/11) of the papers accepted in the CS track in 2019 used Quality-Diversity optimization and 20% (3/15) in 2018. After the conference, the slides of the tutorial will be hosted on the quality-diversity website (https://quality-diversity.github.io/), which gathers papers and educational content related to quality-diversity algorithms. 

Recent Advances in Particle Swarm Optimization Analysis and Understanding

The main objective of this tutorial will be to inform particle swarm optimization (PSO) practitioners of the many common misconceptions and falsehoods that are actively hindering a practitioner’s successful use of PSO in solving challenging optimization problems. While the behaviour of PSO’s particles has been studied both theoretically and empirically since its inception in 1995, most practitioners unfortunately have not utilized these studies to guide their use of PSO. This tutorial will provide a succinct coverage of common PSO misconceptions, with a detailed explanation of why the misconceptions are in fact false, and how they are negatively impacting results. The tutorial will also provide recent theoretical results about PSO particle behaviour from which the PSO practitioner can now make better and more informed decisions about PSO and in particular make better PSO parameter selections. The tutorial will focus on the following important aspects of PSO behaviour
• Understanding why the random variables used in the velocity update should not be scalars, but rather vectors of random values
• Exploring the effects of different ways in which velocity can be initialized
• Clearing up issues with reference to velocity clamping
• The influence of social topology and different iteration strategies on performance is discussed
• Understanding PSO control parameters, and how to use them more efficiently
o The importance of parameter selection will be illustrated with an interactive demo where audience members
will vote for/suggest control parameters. From the set of audience selected control parameters relative
performance ranking, based on popular benchmark suites, of these configuration will be given in relation to a
very extensive set of possible configurations. This demo will clearly illustrate why the subsequent theoretical discussion on
control parameters is so import for effective PSO use.
• Existing theoretical PSO results and what they mean to a PSO practitioner
• Roaming behaviour of PSO particles
• Understanding why PSO struggles with large-scale optimization problems
• Known stability criteria of PSO algorithms
• Effects of particle stability of PSO’s performance
• How to derive new stability criteria for PSO variants and verify them
o The use of a general PSO stability theorem in addition to a simple to use specialization are demonstrated
o The newly derived specialization theorem allows for a non-restrictive relationship of PSO control coefficients, which is a first for
PSO stability theory
• Control parameter tuning, and self-adaptive control parameters
• Is the PSO a local optimizer or a global optimizer?

With the knowledge presented in this tutorial a PSO practitioner will gain up to date theoretical insights into PSO behaviour and as a result be able to make informed choices when utilizing PSO.

NEW Replicability and Reproducibility in Evolutionary Optimization (cancelled)

The reproducibility of experimental results is one of the corner-stones of experimental science, yet often, published experimental results are neither replicable nor reproducible due to insufficient details, missing datasets or software implementations. The Association for Computing Machinery (ACM) distinguishes among “repeatability”, “replicability” and “reproducibility” and it has instituted different badges to be attached to research articles in ACM publications depending on the level of reproducibility. The new ACM journal “Transactions on Evolutionary Learning and Optimization (TELO)” uses these badges to encourage the publication of reproducible results.

The tutorial will be structured in two main parts. The first part will introduce basic concepts in reproducible research, the motivation behind it, and potential pitfalls illustrated from real-world examples. This part will also explain in detail the ACM standards for reproducibility and explain how the process runs at the ACM TELO journal. The second part will describe several techniques for improving reproducibility, ranging from trivial but surprisingly effective to somewhat more technical and laborious. Techniques presented will be demonstrated during the session step-by-step.

Semantic Genetic Programming

Semantic genetic programming is a rapidly growing trend in Genetic Programming (GP) that aims at opening the ‘black box’ of the evaluation function and make explicit use of more information on program behavior in the search. In the most common scenario of evaluating a GP program on a set of input-output examples (fitness cases), the semantic approach characterizes program with a vector of outputs rather than a single scalar value (fitness). The past research on semantic GP has demonstrated that the additional information obtained in this way facilitates designing more effective search operators. In particular, exploiting the geometric properties of the resulting semantic space leads to search operators with attractive properties, which have provably better theoretical characteristics than conventional GP operators. This in turn leads to dramatic improvements in experimental comparisons.

The aim of the tutorial is to give a comprehensive overview of semantic methods in genetic programming, illustrate in an accessible way the formal geometric framework for program semantics to design provably good mutation and crossover operators for traditional GP problem domains, and to analyze rigorously their performance (runtime analysis). A recent extension of this framework to Grammatical Evolution will be also presented. Other promising emerging approaches to semantics in GP will be reviewed. In particular, the recent developments in the behavioural programming and approaches that automatically acquire a multi-objective characterization of programs will be covered as well. Current challenges and future trends in semantic GP will be identified and discussed.

Sequential Experimentation By Evolutionary Algorithms

This tutorial addresses applications of Evolutionary Algorithms (EAs) to global optimization tasks where the objective function cannot be calculated (no explicit model nor a simulation exist), but rather requires a measurement/assay ("wet experiment") in the real-world – e.g., in pharmaceuticals, biocatalyst design, protein expression, quantum processes – to mention only a few.
The use of EAs for experimental optimization is placed in its historical context with an overview of the landmark studies in this area carried out in the 1960s at the Technical University of Berlin. At the same time, statistics-based Design of Experiments (DoE) methodologies, rooted in the late 1950s, constitute a gold-standard in existing laboratory equipment, and are therefore reviewed as well at an introductory level to EC audience.
The main characteristics of experimental optimization work, in comparison to optimization of simulated systems, are discussed, and practical guidelines for real-world experiments with EAs are given. For example, experimental problems can constrain the evolution due to overhead considerations, interruptions, changes of variables, missing assays, imposed population-sizes, and target assays that have different evaluation times (in the case of multiple objective optimization problems).
Selected modern-day case studies show the persistence of experimental optimization problems today. These cover experimental quantum systems, combinatorial drug discovery, protein expression, and others. These applications can throw EAs out of their normal operating envelope, and raise research questions in a number of different areas ranging across constrained EAs, multiple objective EAs, robust and reliable methods for noisy and dynamic problems, and metamodeling methods for expensive evaluations.

Solving Complex Problems with Coevolutionary Algorithms

Coevolutionary algorithms (CoEAs) go beyond the conventional paradigm of evolutionary computation in having the potential to answer questions about what to evolve against (competition) and / or establish how multi-agent behaviours can be discovered (cooperation). Competitive coevolution can be considered from the perspective of discovering tests that distinguish between the performance of candidate solutions. Cooperative coevolution implies that mechanisms are adopted for distributing fitness across more than one individual. In both these variants, the evolving entities engage in interactions that affect all the engaged parties, and result in search gradients that may be very different from those observed in conventional evolutionary algorithm, where fitness is defined externally. This allows CoEAs to model complex systems and solve problems that are difficult or not naturally addressed using conventional evolution.

This tutorial will begin by first establishing basic frameworks for competitive and cooperative coevolution and noting the links to related formalisms (interactive domains and test-based problems). We will identify the pathologies that potentially appear when assuming such coevolutionary formulations (disengagement, forgetting, mediocre stable states) and present the methods that address these issues. Compositional formulations will be considered in which hierarchies of development are explicitly formed leading to the incremental complexification of solutions. The role of system dynamics will also be reviewed with regards to providing additional insight into how design decisions regarding, say, the formulation assumed for cooperation, impact on the development of effective solutions. We will also present the concepts of coordinate systems and underlying objectives and how they can make search/learning more effective. Other covered developments will include hybridization with local search and relationships to shaping.

NEW Statistical Analyses for Meta-heuristic Stochastic Optimization Algorithms

Moving to the era of explainable AI, a comprehensive comparison of the performance of stochastic optimization algorithms has become increasingly important task. One of the most common ways to compare the performance of stochastic optimization algorithms is to apply statistical analyses. However, for performing them, there are still caveats that need to be addressed for acquiring relevant and valid conclusions. First of all, such statistical analyses require good knowledge from the user to apply them properly, which is often lacking and leads to incorrect conclusions. Secondly, the standard approaches can be influenced by outliers (e.g., poor runs) or some statistically insignificant differences (solutions within some ε-neighborhood) that exist in the data.
This tutorial will provide an overview of the current approaches for analyzing algorithms performance with special emphasis on caveats that are often overlooked. We will show how these can be easily avoided by applying simple principles that lead to Deep Statistical Comparison. The tutorial will not be based on equations, but mainly examples through which a deeper understanding of statistics will be achieved. Examples will be based on various comparisons scenarios including single- and multi-objective optimization algorithms. The tutorial will end with a demonstration of a web-service-based framework for statistical comparison of stochastic optimization algorithms.

NEW Theory and Practice of Population Diversity in Evolutionary Computation

Divergence of character is a cornerstone of natural evolution. On the contrary, evolutionary optimization processes are plagued by an endemic lack of population diversity: all candidate solutions eventually crowd the very same areas in the search space. The problem is usually labeled with the oxymoron “premature convergence” and has very different consequences on the different applications, almost all deleterious. At the same time, case studies from theoretical runtime analyses irrefutably demonstrate the benefits of diversity.

This tutorial will give an introduction into the area of “diversity promotion”: we will define the term “diversity” in the context of Evolutionary Computation, showing how practitioners tried, with mixed results, to promote it.

Then, we will analyze the benefits brought by population diversity in specific contexts, namely global exploration and enhancing the power of crossover. To this end, we will survey recent results from rigorous runtime analysis on selected problems. The presented analyses rigorously quantify the performance of evolutionary algorithms in the light of population diversity, laying the foundation for a rigorous understanding of how search dynamics are affected by the presence or absence of diversity and the introduction of diversity mechanisms.

Visualization in Multiobjective Optimization

Visualization in evolutionary multiobjective optimization is relevant in many aspects, such as estimating the location, range, and shape of a solution set approximating the Pareto front (known as an approximation set), assessing conflicts and trade-offs between objectives, selecting preferred solutions, monitoring the progress or convergence of an optimization run, assessing the relative performance of different algorithms, and problem understanding through landscape analysis.

This tutorial will provide an overview of representative methods used in multiobjective optimization for visualizing: (1) individual approximation sets resulting from a single algorithm run, (2) multiple approximation sets stemming from repeated runs, and (3) multiobjective problem landscapes. The methods will be organized according to our recently proposed taxonomy that builds on the nature of the visualized data and the properties of visualization methods.

The methods for visualizing approximation sets will be analyzed according to a methodology that uses a list of requirements for visualization methods and benchmark approximation sets in a similar way as performance metrics and benchmark test problems are used for comparing optimization algorithms. The methods for visualizing multiobjective problem landscapes will be demonstrated on a collection of biobjective problems of increasing difficulty.

NEW Addressing Ethical Challenges within Evolutionary Computation Applications

Robots and artificial intelligence demonstrate to effectively contribute to an increasing number of different domains. At the same time, an increasing number of people – in the general public as well as in research – have started to consider a number of potential ethical challenges related to the development and use of such technology. There are also initiatives across countries like the European Commission appointed High-Level Expert Group on Artificial Intelligence (AI HLEG) that has as a general objective to support the implementation of the European Strategy on Artificial Intelligence (please add link as hidden: https://ec.europa.eu/digital-single-market/en/artificial-intelligence). This talk will give an overview of the most commonly expressed ethical challenges and ways being undertaken to reduce their impact using the findings in an earlier undertaken review (please add link as hidden: https://www.frontiersin.org/articles/10.3389/frobt.2017.00075/full) supplemented with recent work and initiatives.

Among the most important challenges are those related to privacy, safety and security. Countermeasures can be taken first at design time, second, when a user should decide where and when to apply a system and third, when a system is in use in its environment. In the latter case, there will be a need for the system by itself to perform some ethical reasoning if operating in autonomous mode. We are currently undertaking research in various projects where the challenges appear. The tutorial will introduce some examples from our own and others work and how the challenges can be addressed both from a technical and human side with special attention to problems relevant when working with evolutionary computation. Ethical issues should not be seen only as challenges but also as new research opportunities contributing to more useful services and systems.

Automated Algorithm Configuration and Design

Most optimization algorithms, including evolutionary algorithms and
metaheuristics, and general-purpose solvers for integer or constraint
programming, have often many parameters that need to be properly
designed and tuned for obtaining the best results on a particular
problem. Automatic (offline) algorithm design methods help algorithm
users to determine the parameter settings that optimize the
performance of the algorithm before the algorithm is actually
deployed. Moreover, automatic offline algorithm design methods may
potentially lead to a paradigm shift in algorithm design because they
enable algorithm designers to explore much larger design spaces than
by traditional trial-and-error and experimental design
procedures. Thus, algorithm designers can focus on inventing new
algorithmic components, combine them in flexible algorithm frameworks,
and let final algorithm design decisions be taken by automatic
algorithm design techniques for specific application contexts.

This tutorial is structured into two main parts. In the first part, we
will give an overview of the algorithm design and tuning problem,
review recent methods for automatic algorithm design, and illustrate
the potential of these techniques using recent, notable applications
from the presenters' and other researchers work. In the second part of
the tutorial will focus on a detailed discussion of more complex
scenarios, including multi-objective problems, anytime algorithms,
heterogeneous problem instances, and the automatic generation of
algorithms from algorithm frameworks. The focus of this second part of
the tutorial is, hence, on practical but challenging applications of
automatic algorithm design. The second part of the tutorial will
demonstrate how to tackle algorithm design tasks using our irace
software (http://iridia.ulb.ac.be/irace), which implements the
iterated racing procedure for automatic algorithm design. We will
provide a practical step-by-step guide on using irace for the typical
algorithm design scenario.

NEW EA & ML, synergies and challenges

While Machine Learning (ML) techniques enjoyed growing popularity in recent years, the role of Evolutionary Algorithms in this field is still marginal — quite a surprising fact considering how deeply the origins of the two fields are related.

In this tutorial we present success stories of EAs exploited in specific ML tasks, such as feature selection, adversarial ML, whitebox modeling, also mentioning the renowned neuroevolution. We show how similar concepts appear in both fields with different names.

At the same time, we show well-known and emerging challenges that EAs need to overcome to become widely adopted in ML. For instance, a reduced ability to scale or a general distrust toward stochasticity.

Finally, we point out opportunities arising for new research lines, that play on the strengths of EAs, such as potential improvements over currently-used optimization techniques; and the capability to go beyond simple model fitting, creating solutions that expand over the boundaries of the training data.

NEW Evolutionary Algorithms in Biomedical Data Mining: Challenges, Solutions, and Frontiers

Evolutionary algorithms offer an open-ended, nature inspired set of strategies for solving real-world problems that set them apart from other methods found in machine learning and optimization. The domain of biomedical data mining must confront many unique challenges (e.g. data scale, noise, bias, data completeness, complex multivariate associations, heterogeneity, and the need for interpretability). This tutorial will highlight these challenges, paired with solutions that have been proposed and adopted by the research community as well as identify promising cutting-edge evolutionary computation strategies expected to be of benefit to biomedical analysis applications in the future. It will seek to distinguish challenges that are generalizable to the broader field of EC from those that are specific to subfamilies of methodological development and application (e.g. genetic algorithms, genetic programming, co-evolutionary algorithms, learning classifier systems, neuro-evolution, evolutionary programming, differential evolution). Themes for discussion will include, but not be limited to; stochasticity, parameter optimization, representability vs. evolvability, interpretability, evaluation metrics, maintaining diversity, generations and elitism, selection, genetic operators, defining fitness, and confounding the objective with the objective function. These broader EC themes will be punctuated with a variety of specific examples of application within the broader biomedical research domain

Evolutionary Computation and Evolutionary Deep Learning for Image Analysis, Signal Processing and Pattern Recognition

The intertwining disciplines of image analysis, signal processing and pattern recognition are major fields of computer science, computer engineering and electrical and electronic engineering, with past and on-going research covering a full range of topics and tasks, from basic research to a huge number of real-world industrial applications.

Among the techniques studied and applied within these research fields, evolutionary computation (EC) including evolutionary algorithms, swarm intelligence and other paradigms is playing an increasingly relevant role. Recently, evolutionary deep learning has also attracted very good attention to these fields. The terms Evolutionary Image Analysis and Signal Processing and Evolutionary Computer Vision are more and more commonly accepted as descriptors of a clearly defined research area and family of techniques and applications. This has also been favoured by the recent availability of environments for computer hardware and systems such as GPUs and grid/cloud/parallel computing systems, whose architecture and computation paradigm fit EC algorithms extremely well, alleviating the intrinsically heavy computational burden imposed by such techniques and allowing even for real-time applications.

The tutorial will introduce the general framework within which Evolutionary Image Analysis, Signal Processing and Pattern Recognition can be studied and applied, sketching a schematic taxonomy of the field and providing examples of successful real-world applications. The application areas to be covered will include edge detection, segmentation, object tracking, object recognition, motion detection, image classification and recognition. EC techniques to be covered will include genetic algorithms, genetic programming, particle swarm optimisation, evolutionary multi-objective optimisation as well as memetic/hybrid paradigms. In particular, we will discuss the detection of relevant set of features for classification based on an information-theoretical approach derived from complex system analysis. We take a focus on the use of evolutionary deep learning idea for image analysis --- this includes automatic learning architectures, learning parameters and transfer functions of convolutional neural networks (and autoencoders and genetic programming if time allows). The use of GPU boxes will be discussed for real-time/fast object classification. We will show how such EC techniques can be effectively applied to image analysis and signal processing problems and provide promising results.

Evolutionary Computation and Machine Learning in Cryptology

In recent years, the interplay between artificial intelligence (AI) and security is becoming more prominent and important. This comes naturally because of the need to improve security in a more automated way. One specific domain of security that steadily receives more AI applications is cryptology. There, we already see how AI techniques can improve implementation attacks, attacks on PUFs, hardware Trojan detection, etc.
This tutorial first gives a brief introduction into domains where AI (we concentrate on machine learning and evolutionary algorithms) is used to solve problems from the security domain.
Next, we focus on problems coming from the cryptology domain.
We look at several realistic crypto problems successfully tackled with EAs and machine learning and discuss why those problems are suitable to apply such techniques. Some representative examples of the problems we cover are the evolution of Boolean functions and S-boxes, machine learning and EA attacks on Physically Unclonable Functions, machine learning for side-channel attacks, EA/machine learning to improve fault injection, hardware Trojan detection/prevention/insertion, attacks on logic locking, neuroevolution applications, etc.
We finish this tutorial with a discussion about how experiences in solving problems in AI could help to solve problems in security and vice versa. To that end, we emphasize a collaborative approach for both communities.

Evolutionary Computation for Feature Selection and Feature Construction

In data mining/big data and machine learning, many real-world problems such as bio-data classification and biomarker detection, image analysis, text mining often involve a large number of features/attributes. However, not all the features are essential since many of them are redundant or even irrelevant, and the useful features are typically not equally important. Using all the features for classification or other data mining tasks typically does not produce good results due to the big dimensionality and the large search space. This problem can be solved by feature selection to select a small subset of original (relevant) features or feature construction to create a smaller set of high-level features using the original low-level features.

Feature selection and construction are very challenging tasks due to the large search space and feature interaction problems. Exhaustive search for the best feature subset of a given dataset is practically impossible in most situations. A variety of heuristic search techniques have been applied to feature selection and construction, but most of the existing methods still suffer from stagnation in local optima and/or high computational cost. Due to the global search potential and heuristic guidelines, evolutionary computation techniques such as genetic algorithms, genetic programming, particle swarm optimisation, ant colony optimisation, differential evolution and evolutionary multi-objective optimisation have been recently used for feature selection and construction for dimensionality reduction, and achieved great success. Many of these methods only select/construct a small number of important features, produce higher accuracy, and generated small models that are easy to understand/interpret and efficient to run on unseen data. Evolutionary computation techniques have now become an important means for handling big dimensionality issues where feature selection and construction are required.

The tutorial will introduce the general framework within which evolutionary feature selection and construction can be studied and applied, sketching a schematic taxonomy of the field and providing examples of successful real-world applications. The application areas to be covered will include bio-data classification and biomarker detection, image analysis and pattern classification, symbolic regression, network security and intrusion detection, and text mining. EC techniques to be covered will include genetic algorithms, genetic programming, particle swarm optimisation, differential evolution, ant colony optimisation, artificial bee colony optimisation, and evolutionary multi-objective optimisation. We will show how such evolutionary computation techniques (with a focus on particle swarm optimisation and genetic programming) can be effectively applied to feature selection/construction and dimensionality reduction and provide promising results.

Evolutionary Computer Vision

This tutorial, based on the book "Evolutionary Computer Vision," published with Springer in the Natural Computing series, presents the subject under the umbrella of goal-driven vision. Here the author explains the theory and application of evolutionary computer vision, a new paradigm where challenging vision problems can be approached using the techniques of evolutionary computing. This methodology achieves excellent results for defining fitness functions and representations for problems by merging evolutionary computation with mathematical optimization to produce automatic creation of emerging visual behaviors.
In the first part of the tutorial, the author surveys the literature in a concise form, defines the relevant terminology, and offers historical and philosophical motivations for the fundamental research problems in the field. The second part of the tutorial focuses on implementing evolutionary algorithms that solve given problems using working programs in the primary areas of low-, intermediate- and high-level computer vision.

NEW Multi-concept Optimization (cancelled)

The main goal is to introduce Multi-Concept Optimization (MCO) to the GECCO community of researches and practitioners. This goal will be achieved by way of the following seven tasks:
1. Providing background on what is a conceptual solution (as evident from the conceptual design stage during the process of engineering design)
2. Describing academic and real-life examples of conceptual solutions
3. Defining what is MCO and how it differs from a traditional definition of an optimization problem. The definitions will include both single and multi-objective MCO
4. Explaining the significance of MCO as an optimization methodology that is useful for the following three main reasons:
a. Supporting the selection of conceptual solutions (e.g., 1-3)
b. An alternative approach to multi-modal optimization (see explanation in 4)
c. A unique approach to design space exploration (e.g., 5)
5. Describing evolutionary algorithms for MCO and their benchmarking (e.g., 3, 5-6)
6. Describing the application of MCO to a real-life problem (joint work with Israel Aerospace Industries)
7. Discussing the research needs concerning evolutionary computation for MCO
Indicative Bibliography:
1 Mattson C. A., Messac, A. Pareto frontier based concept selection under uncertainty with visualization. Opt. and Eng., 6; p. 85–115, 2005.
2 Avigad, G., and Moshaiov, A. Set-based concept selection in multi-objective problems: optimality versus variability approach. J. of Eng. Design, Vol. 20, No. 3, pp. 217-242, 2009.
3 Avigad, G. and Moshaiov, A. Interactive evolutionary multiobjective search and optimization of set-based concepts. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 39(4), pp.1013-1027, 2009.
4 Moshaiov, A. The Paradox of Multimodal Optimization: Concepts vs. Species in Single and Multi-objective Problems. Proceedings of the IEEE Congress on Evolutionary Computation, 2016.
5 Moshaiov, A., Snir, A., and Samina, B. Concept-based evolutionary exploration of design spaces by a resolution-relaxation-Pareto approach, Proceedings of the IEEE Congress on Evolutionary Computation, pp. 1845-1852, 2015.
6 Farhi, E. and Moshaiov, A. Window-of-Interest-based Multi-objective Evolutionary Search for Satisficing Concepts. Proceedings of the IEEE Conference on Systems, Man and Cybernetics, 2017.

Push

Push is a general purpose, multi-type, Turing complete programming language for programs that evolve in an evolutionary computation system. Initially developed in 2000, it has been used for a range of research projects and applications, including the production of human-competitive results and state-of-the-art work on general program synthesis.

Push and systems that evolve Push programs (such as PushGP) are available in a variety of host languages. However, the core concepts of Push are simple enough that programmers should also be able to build their own Push systems relatively quickly, and to incorporate them into their own evolutionary computation systems. This tutorial will present examples and best practices that will help participants to use Push effectively in their own work, whether they use existing libraries or develop their own.

Push has unusually simple syntax, which makes it particularly easy to randomly generate and vary valid programs. It nonetheless supports rich data and control structures, which in most other languages involve syntax restrictions that can be difficult to maintain during evolutionary change. Furthermore, the data and control structures used by a Push program can themselves emerge through the evolutionary process.

This tutorial will provide a detailed introduction to the Push programming language, and will demonstrate the use of the PushGP genetic programming system with open source libraries in Python (PyshGP) and Clojure (Clojush). Participants will be invited to install and interact with these libraries during the tutorial.

Among the features of Push that will be illustrated are those that support the evolution of programs that use multiple types, iteration, recursion, and modularity to solve problems that may involve multiple tasks. Push-based "autoconstructive evolution" systems, in which evolutionary methods co-evolve with solutions to problems, will also be briefly described.

NEW Search Based Software Engineering: challenges, opportunities and recent applications

Modern software development has entered a mass production era with a multitude of engineering challenges in practice to provide higher performance and better functionality. To address these challenges, a growing trend has begun in recent years to move software engineering problems from human-based search to machine-based search that balances a number of constraints to achieve optimal or near-optimal solutions. As a result, human effort is moving up the abstraction chain to focus on guiding the automated search, rather than performing the search itself. This emerging software engineering paradigm is known as Search Based Software Engineering (SBSE). It uses search based optimization techniques, mainly those from the evolutionary computation literature to automate the search for optimal or near-optimal solutions to software engineering problems. The SBSE approach can and has been applied to many problems in software engineering that span the spectrum of activities along the software lifecycle from requirements to maintenance and reengineering. Already, success has been achieved in requirements, refactoring, project planning, testing, maintenance and reverse engineering. However, several challenges have to be addressed to mainly tackle the growing complexity of software systems nowadays in terms of number of objectives, constraints and inputs/outputs.

In this tutorial, we will cover the basic knowledge about SBSE and explain its challenges and opportunities. Then, we will focus on recent case studies that we proposed, along with our research group and industrial partners, addressing the above challenges. These case studies will cover a range of topics in software engineering including: multi-objective software design defects detection, static and interactive multi-objective optimization for software refactoring, and multi-objective optimization for third-party software components reuse. Finally, we will discuss possible new research directions in SBSE.

NEW Swarm Intelligence in Cybersecurity

Information play in today's life crucial role. Thus, the topic of the cybersecurity and especially malicious code (also known as malware) become to be more important today, since such malicious codes are not directed only at the computer systems, but also at other electronic devices that may be open to attack and interfere with our private lives and contain personal or valuable data. Currently, Artificial Intelligence (AI) started to play a significant role in cybersecurity threads identification and defense. Evolutionary algorithms and swarm intelligence belonging to the AI paradigm has been successfully and widely used for solving various real-world problems. One domain with many open problems is cybersecurity. Recent developments in AI techniques point to a high potential for both cybersecurity defenders and cybercriminals. Antimalware solutions may utilize intelligent techniques to detect and prevent cyber threats, and at the same time, these intelligent techniques can be used for cybercriminal activities. This tutorial bridge the gap between the cybersecurity and AI community with the special emphasis on how intelligent techniques can be used to create and upgrade concepts of malicious code as a background for the future effective antimalware solutions. The tutorial covers two main parts.
The tutorial starts with a brief introduction to cybersecurity, including online thread monitoring demonstrations, general concepts/principles of viruses, malware, and cybernetic weapons (e.g., Stuxnet). The second part examines the mutual fusion of swarm intelligence/algorithms and both malware and antimalware as the very likely possible near future threads. We will discuss the choice of appropriate swarm intelligence techniques, visualization methods, and analyses for various case studies and evaluate the importance of that choice. The tutorial finish with a general overview of the development in the interconnected cybersecurity and AI fields, experiences with real-world experiments from our lab, live demos, exhibiting experimental swarm bot, and future concept of swarm antimalware.
The tutorial is based on the most actual state of the art, as well as on our original research and experiments published in various journals, leading conferences, and books.
The tutorial is designed for GECCO general audience; advanced or expert knowledge of cybersecurity is not expected.

NEW Theory of Estimation-of-Distribution Algorithms

Estimation-of-distribution algorithms (EDAs) are general
metaheuristics for optimization that represent a more recent and
popular alternative to classical approaches like evolutionary
algorithms. In a nutshell, EDAs typically do not directly evolve
populations of search points but build probabilistic models of
promising solutions by repeatedly sampling and selecting points from
the underlying search space. However, until recently the theoretical
knowledge of the working principles of EDAs was relatively limited.

In the last few years, there has been made significant progress in the
theoretical understanding of EDAs. This tutorial provides an
up-to-date overview of the most commonly analyzed EDAs and the most
important theoretical results in this area, ranging from convergence
to runtime analyses. Different algorithms and objective functions,
including optimization under uncertainty (i. e. noise) are
covered. The tutorial will present typical benchmark functions and
tools relevant in the theoretical analysis. It will be demonstrated
that some EDAs are very sensitive with respect to their parameter
settings and that their runtime can be optimized for very different
choices of learning parameters. Altogether, the tutorial with make
the audience familiar with the state of the art in the theory of EDAs
and the most useful techniques for the analysis. We will conclude with
open problems and directions for future research.