EuroGP is the premier annual conference on Genetic Programming, the oldest and the only meeting worldwide devoted specifically to this branch of evolutionary computation. It is always a very enjoyable event attracting participants from all continents, offering excellent opportunities for networking, informal contact, exchange of ideas and discussions with fellow researchers, in a friendly and relaxed environment. High quality papers describing new original research are sought on topics strongly related to the evolution of computer programs, ranging from theoretical work to innovative applications. The conference will feature a mixture of oral presentations and poster sessions.
Topics include but are not limited to:
A comprehensive bibliography of genetic programming literature and links to related material is accessible at the Genetic Programming Bibliography web page, part of the Collection of Computer Science Bibliographies maintained and managed by William Langdon, Steven Gustafson, and John Koza.
All accepted papers will be printed in the proceedings published by Springer Verlag in the Lecture Notes in Computer Science (LNCS) series.
Previous editions of EuroGP were published in the following Springer Verlag LNCS volumes
The papers which receive the best reviews will be nominated for the Best Paper Award.
Submissions must be original and not published elsewhere. They will be peer reviewed by at least three members of the program committee. The reviewing process will be double-blind, so please omit information about the authors in the submitted paper. Submit your manuscript in Springer LNCS format.
EuroGP accepts two kind of submissions: full papers and short papers. Full papers require novel and complete research work and have a limit of 16 pages. Short papers should present complete research or interesting preliminary results and have a limit of 8 pages. Both types of submission will undergo the same blind review process and all accepted papers will be included in the LNCS proceedings.
Authors of accepted papers will have to improve their paper on the basis of the reviewers' comments and will be asked to send a camera ready version of their manuscripts. At least one author of each accepted work has to register for the conference, attend the conference and present the work.
Submission link: https://myreview.saclay.inria.fr/eurogp18
Title: A Multiple Expression Alignment Framework for Genetic Programming
Author: Leonardo Vanneschi, Kristen Scott, Mauro Castelli
Abstract: Alignment in the error space is a recent idea to exploit semantic awareness in genetic programming. In a previous contribution, the concepts of optimally aligned and optimally coplanar individuals were introduced, and it was shown that given optimally aligned, or optimally coplanar, individuals, it is possible to construct a globally optimal solution analytically. As a consequence, genetic programming methods, aimed at searching for optimally aligned, or optimally coplanar, individuals were introduced. In this paper, we critically discuss those methods, analyzing their major limitations and we propose new genetic programming systems aimed at overcoming those limitations. The presented experimental results, conducted on five real-life symbolic regression problems, show that the proposed algorithms outperform not only the existing methods based on the concept of alignment in the error space, but also geometric semantic genetic programming and standard genetic programming.
Title: Evolving Graphs by Graph Programming
Author: Timothy Atkinson, Detlef Plump, Susan Stepney
Abstract: Rule-based graph programming is a deep and rich topic. We present an approach to exploiting the power of graph programming as a representation and as an execution medium in an evolutionary algorithm (EGGP). We demonstrate this power in comparison with Cartesian Genetic Programming (CGP), showing that it is significantly more efficient in terms of fitness evaluations on some classic benchmark problems. We hypothesise that this is due to its ability to exploit the full graph structure, leading to a richer mutation set, and outline future work to test this hypothesis, and to exploit further the power of graph programming within an EA.
Title: Pruning Techniques for Mixed Ensembles of Genetic Programming Models
Author: Mauro Castelli, Ivo Gonçalves, Luca Manzoni, Leonardo Vanneschi
Abstract: The objective of this paper is to define an effective strategy for building an ensemble of Genetic Programming (GP) models. Ensemble methods are widely used in machine learning due to their features: they average out biases, they reduce the variance and they usually generalize better than single models. Despite these advantages, building ensemble of GP models is not a well-developed topic in the evolutionary computation community. To fill this gap, we propose a strategy that blends individuals produced by standard syntax-based GP and individuals produced by geometric semantic genetic programming, one of the newest semantics-based method developed in GP. In fact, recent literature showed that combining syntactic and semantics could improve the generalization ability of a GP model. Additionally, to improve the diversity of the GP models used to build up the ensemble, we propose different pruning criteria that are based on correlation and entropy, a commonly used measure in information theory. Experimental results, obtained over different complex problems, suggest that the pruning criteria based on correlation and entropy could be effective in improving the generalization ability of the ensemble model and in reducing the computational burden required to build it.
Title: Generating Redundant Features with Unsupervised Multi-Tree Genetic Programming
Author: Andrew Lensen, Bing Xue, Mengjie Zhang
Abstract: Recently, feature selection has become an increasingly important area of research due to the surge in high-dimensional datasets in all areas of modern life. A plethora of feature selection algorithms have been proposed, but it is difficult to truly analyse the quality of a given algorithm. Ideally, an algorithm would be evaluated by measuring how well it removes known bad features. Acquiring datasets with such features is inherently difficult, and so a common technique is to add synthetic bad features to an existing dataset. While adding noisy features is an easy task, it is very difficult to automatically add complex, redundant features. This work proposes one of the first approaches to generating redundant features, using a novel genetic programming approach. Initial experiments show that our proposed method can automatically create difficult, redundant features which have the potential to be used for creating high-quality feature selection benchmark datasets.
Title: Towards In Vivo Genetic Programming: Evolving Boolean Networks to Determine Cell States
Author: Nadia Taou, Michael Lones
Abstract: Within the genetic programming community, there has been growing interest in the use of computational representations motivated by gene regulatory networks (GRNs). It is thought that these representations capture useful biological properties, such as evolvability and robustness, and thereby support the evolution of complex computational behaviours. However, computational evolution of GRNs also opens up opportunities to go in the opposite direction: designing programs that could one day be implemented in biological cells. In this paper, we explore the ability of evolutionary algorithms to design Boolean networks, abstract models of GRNs suitable for refining into synthetic biology implementations, and show how they can be used to control cell states within a range of executable models of biological systems.
Title: Evolving the Topology of Large Scale Deep Neural Networks
Author: Filipe Assunção, Nuno Lourenço, Penousal Machado, Bernardete Ribeiro
Abstract: In the recent years Deep Learning has attracted a lot of attention due to its success in difficult tasks such as image recognition and computer vision. Most of the success in these tasks is merit of Convolutional Neural Networks (CNNs), which allow the automatic construction of features. However, designing such networks is not an easy task, which requires expertise and insight. In this paper we introduce DENSER, a novel representation for the evolution of deep neural networks. In concrete we adapt ideas from Genetic Algorithms (GAs) and Grammatical Evolution (GE) to enable the evolution of sequences of layers and their parameters. We test our approach in the well-known image classification CIFAR-10 dataset. The results show that our method: (i) outperforms previous evolutionary approaches to the generations of CNNs; (ii) is able to create CNNs that have state-of-the-art performance while using less prior knowledge (iii) evolves CNNs with novel topologies, unlikely to be designed by hand. For instance, the best performing CNNs obtained during evolution has an unexpected structure using six consecutive dense layers. On the CIFAR-10 the best model reports an average error of 5.87% on test data.
Title: Using GP is NEAT: Evolving Compositional Pattern Production Functions
Author: Filipe Assunção, Nuno Lourenço, Penousal Machado, Bernardete Ribeiro
Abstract: The success of Artificial Neural Networks (ANNs) highly depends on their architecture and on how they are trained. However, making decisions regarding such domain specific issues is not an easy task, and is usually performed by hand, through an exhaustive trial-and-error process. Over the years, researches have developed and proposed methods to automatically train ANNs. One example is the HyperNEAT algorithm, which relies on NeuroEvolution of Augmenting Topologies (NEAT) to create Compositional Pattern Production Networks (CPPNs). CPPNs are networks that encode the mapping between neuron positions and the synaptic weight of the ANNs connection between those neurons. Although this approach has obtained some success, it requires meticulous parameterisation to work properly. In this article we present a comparison of different Evolutionary Computation methods to evolve Compositional Pattern Production Functions: structures that have the same goal as CPPNs, but that are encoded as functions instead of networks. In addition to NEAT three methods are used to evolve such functions: Genetic Programming (GP), Grammatical Evolution, and Dynamic Structured Grammatical Evolution. The results show that GP is able to obtain competitive performance, often surpassing the other methods, without requiring the fine tuning of the parameters.
Title: Structurally Layered Representation Learning: Towards Deep Learning through Genetic Programming
Author: Lino Rodriguez-Coayahuitl, Alicia Morales-Reyes, Hugo Jair Escalante
Abstract: We introduce a novel method for representation learning based on genetic programming (GP). Inspired into the way that deep neural networks learn descriptive/discriminative representations from raw data, we propose a structurally layered representation that allows GP to learn a feature space from large scale and high dimensional data sets. Previous efforts from the GP community for feature learning have focused on small data sets with a few input variables, also, most approaches rely on domain expert knowledge to produce useful representations. In this paper, we introduce the structurally layered GP formulation, together with an efficient scheme to explore the search space and show that this framework can be used to learn representations from large data sets of high dimensional raw data. As case of study we describe the implementation and experimental evaluation of an autoencoder developed under the proposed framework. Results evidence the benefits of the proposed framework and pave the way for the development of deep genetic programming.
Title: Evolving Better RNAfold Structure Prediction
Author: Justyna Petke, Ronny Lorenz
Abstract: Grow and graft genetic programming (GGGP) evolves more than 50000 parameters in a state-of-the-art C program to make functional source code changes which give more accurate predictions of how RNA molecules fold up. Genetic improvement updates 29percent of the dynamic programming free energy model parameters. In most cases (50.3percent) GI gives better predictions on 4655 known secondary structures from RNA\_STRAND (29.0percent are worse and 20.7percent are unchanged, p = 0.
Title: Genetic Programming Hyper-heuristic with Cooperative Coevolution for Dynamic Flexible Job Shop Scheduling
Author: Daniel Yska, Yi Mei, Mengjie Zhang
Abstract: Flexible Job Shop Scheduling (FJSS) problem has many real-world applications such as manufacturing and cloud computing, and thus is an important area of study. In real world, the environment is often dynamic, and unpredicted job orders can arrive in real time. Dynamic FJSS consists of challenges of both dynamic optimisation and the FJSS problem. In Dynamic FJSS, two kinds of decisions (so-called routing and sequencing decisions) are to be made in real time. Dispatching rules have been demonstrated to be effective for dynamic scheduling due to their low computational complexity and ability to make real-time decisions. However, it is time consuming and strenuous to design effective dispatching rules manually due to the complex interactions between job shop attributes. Genetic Programming Hyper-heuristic (GPHH) has shown success in automatically designing dispatching rules which are much better than the manually designed ones. Previous works only focused on standard job shop scheduling with only the sequencing decisions. For FJSS, the routing rule is set arbitrarily by intuition. In this paper, we explore the possibility of evolving both routing and sequencing rules together and propose a new GPHH algorithm with Cooperative Co-evolution. Our results show that co-evolving the two rules together can lead to much more promising results than evolving the sequencing rule only.
Title: Geometric Crossover in Syntactic Space
Author: João Macedo, Carlos Fonseca, Ernesto Costa
Abstract: This paper presents a geometric crossover operator for Tree-Based Genetic Programming that acts on the syntactic space, where each expression tree is represented in prefix notation. The proposed operator is compared to the standard subtree crossover on a symbolic regression problem, on the Santa Fe Ant Trail and on a classification problem. Statistically validated results show that the individuals produced using this method are significantly smaller than those produced by the subtree crossover, and have similar or better performance in the target tasks.
Title: A Comparative Study on Crossover in Cartesian Genetic Programming
Author: Jakub Husa, Roman Kalkreuth
Abstract: Cartesian Genetic Programming is often used with mutation as the sole genetic operator. Compared to the comprehensive and detailed knowledge about the effect and use of mutation in CGP, the use of crossover has been less investigated and studied. In this paper, we present a comparative study of previously proposed crossover techniques for Cartesian Genetic Programming. This work also includes the proposal of a new crossover technique which swaps block of the CGP phenotype between two selected parents. The experiments of our study open a new perspective on comparative studies on crossover in CGP and its challenges. Our results show that it is possible for a crossover operator to outperform the standard (1+lambda) strategy on a limited number of tasks. The question of finding a universal crossover operator in CGP remains open.
Title: Comparing Rule Evaluation Metrics for the Evolutionary Discovery of Multi-Relational Association Rules in the Semantic Web
Author: Minh Duc Tran, Claudia d'Amato, Binh Thanh Nguyen, Andrea G. B. Tettamanzi
Abstract: We carry out a comparison of popular asymmetric metrics, originally proposed for scoring association rules, as building blocks for a fitness function for evolutionary inductive programming. In particular, we use them to score candidate multi-relational association rules in an evolutionary approach to the enrichment of populated knowledge bases in the context of the Semantic Web. The evolutionary algorithm searches for hidden knowledge patterns, in the form of SWRL rules, in assertional data, while exploiting the deductive capabilities of ontologies. Our methodology is to compare the number of generated rules and total predictions when the metrics are used to compute the fitness function of the evolutionary algorithm. This comparison, which has been carried out on three publicly available ontologies, is a crucial step towards the selection of suitable metrics to score multi-relational association rules that are generated from ontologies.
Title: Investigating a Machine Breakdown Genetic Programming Approach for Dynamic Job Shop Scheduling
Author: John Park, Yi Mei, Su Nguyen, Gang Chen, Mengjie Zhang
Abstract: Dynamic job shop scheduling (JSS) problems with dynamic job arrivals have been studied extensively in the literature due to their applicability to real-world manufacturing systems, such as semiconductor manufacturing. In a dynamic JSS problem with dynamic job arrivals, jobs arrive on the shop floor unannounced that need to be processed by the machines on the shop floor. A job has a sequence of operations that can only processed on specific machines, and machines can only process one job at a time. Many effective genetic programming based hyper-heuristic (GP-HH) approaches have been proposed for dynamic JSS problems with dynamic job arrivals, where high quality dispatching rules are automatically evolved by GP to handle the dynamic JSS problem instances. However, research that focus on handling multiple dynamic events simultaneously are limited, such as both dynamic job arrivals and machine breakdowns. A machine breakdown event results in the affected machine being unable to process any jobs during the repair time. It is likely that machine breakdowns can significantly affect the effectiveness of the scheduling procedure unless they are explicitly accounted for. Therefore, this paper develops new machine breakdown terminals for a GP approach and evaluates their effectiveness for a dynamic JSS problem with both dynamic job arrivals and machine breakdowns. The results show that the GP approaches with the machine breakdown terminals do show improvements. The analysis shows that the machine breakdown terminals may indirectly contribute in the evolution of high quality rules, but occur infrequently in the output rules evolved by the machine breakdown GP approaches.
Title: Multi-Objective Evolution of Ultra-Fast General-Purpose Hash Functions
Author: David Grochol, Lukas Sekanina
Abstract: Hashing is an important function in many applications such as hash tables, caches and Bloom filters. In past, genetic programming was applied to evolve application-specific as well as general-purpose hash functions, where the main design target was the quality of hashing. As hash functions are frequently called in various time-critical applications, it is important to optimize their implementation with respect to the execution time. In this paper, linear genetic programming is combined with NSGA-II algorithm in order to obtain general-purpose, ultra-fast and high-quality hash functions. Evolved hash functions show highly competitive quality of hashing, but significantly reduced execution time in comparison with the state of the art hash functions available in literature.
Title: Analyzing Feature Importance for Metabolomics using Genetic Programming
Author: Ting Hu, Karoliina Oksanen, Weidong Zhang, Edward Randell, Andrew Furey, Guangju Zhai
Abstract: The emerging and fast-developing field of metabolomics ex- amines the abundance of small-molecule metabolites in body fluids to study the cellular processes related to how the human body responds to genetic and environmental perturbations. Considering the complexity of metabolism, metabolites and their represented cellular processes can correlate and synergistically contribute to a phenotypic status. Genetic programming (GP) provides advanced analytical instruments for the investigation of multifactorial causes of metabolic diseases. In this article, we analyzed a population-based metabolomics dataset on osteoarthritis (OA) and developed a Linear GP (LGP) algorithm to search classification models that can best predict the disease outcome, as well as to identify the most important metabolic markers associated with the dis- ease. The LGP algorithm was able to evolve prediction models with high accuracies especially with a more focused search using a reduced feature set that only includes potentially relevant metabolites. We also identified a set of key metabolic markers that may improve our understanding of the biochemistry and pathogenesis of the disease.
Title: On the Automatic Design of a Representation for Grammar-based Genetic Programming
Author: Eric Medvet, Alberto Bartoli
Abstract: A long-standing problem in Evolutionary Computation consists in how to choose an appropriate representation for the solutions. In this work we investigate the feasibility of synthesizing a representation automatically, for the large class of problems whose solution spaces can be defined by a context-free grammar. We propose a framework based on a form of meta-evolution in which individuals are candidate representations expressed with an ad hoc language that we have developed to this purpose. Individuals compete and evolve according to an evolutionary search aimed at optimizing such representation properties as redundancy, locality, uniformity of redundancy. We assessed experimentally three variants of our framework on established benchmark problems and compared the resulting representations to human-designed representations commonly used (e.g., classical Grammatical Evolution). The results are promising in the sense that the evolved representations indeed exhibit better properties than the human-designed ones. Furthermore, while those improved properties do not result in a systematic improvement of search effectiveness, some of the evolved representations do improve search effectiveness over the human-designed baseline.
Title: Scaling Tangled Program Graphs to Visual Reinforcement Learning in ViZDoom
Author: Robert Smith, Malcolm Heywood
Abstract: A tangled program graph framework (TPG) was recently proposed as an emergent process for decomposing tasks and simultaneously composing solutions by organizing code into graphs of teams of programs. The initial evaluation assessed the ability of TPG to discover agents capable of playing Atari game titles under the Arcade Learning Environment. This is an example of 'visual' reinforcement learning, i.e. agents are evolved directly from the frame buffer without recourse to hand designed features. TPG was able to evolve solutions competitive with state-of-the-art deep reinforcement learning solutions, but at a fraction of the complexity. One simplification assumed was that the visual input could be down sampled from a 210 x 160 resolution to 42 x 32. In this work, we consider the challenging 3D first person shooter environment of ViZDoom and require that agents be evolved at the original visual resolution of 320 x 240 pixels. To do so, we address issues with task scenarios performing fitness evaluation over multiple tasks. The resulting TPG solutions retain all the emergent properties of the original work as well as the computational efficiency. Moreover, solutions appear to generalize across multiple task scenarios, whereas equivalent solutions from deep reinforcement learning have focused on single task scenarios alone.
Title: Multi-Level Grammar Genetic Programming for Scheduling in Heterogeneous Networks
Author: Takfarinas Saber, David Fagan, David Lynch, Stepan Kucera, Holger Claussen, Michael O'Neill
Abstract: Co-ordination of Inter-Cell Interference through scheduling enables telecommunication companies to better exploit their Heterogeneous Networks. However, it requires from these entities to implement an effective scheduling algorithm. The state-of-the-art for the scheduling in Heterogeneous Networks is a Grammar-Guided Genetic Programming algorithm which evolves, from a given grammar, an expression that maps to the scheduling of transmissions. We evaluate in our work the possibility of improving the results obtained by the state-of-the-art using a layered grammar approach. We show that starting with a small restricted grammar and introducing the full functionality after 10 generations outperforms the state-of-the-art, even when varying the algorithm used to generate the initial population and the maximum initial tree depth.