Search

We describe a novel method for using Genetic Programming to create compact classification rules based on combinations of N-Grams (character strings). Genetic programs acquire fitness by producing rules that are effective classifiers in terms of precision and recall when evaluated against a set of training documents. We describe a set of functions and terminals and provide results from a classification task using the Reuters 21578 dataset. We also suggest that because the induced rules are meaningful to a human analyst they may have a number of other uses beyond classification and provide a basis for text mining applications. [ABSTRACT FROM AUTHOR]

We examine a model genetic system that has features of both genetic programming and genetic regulatory networks, to show how various forms of degeneracy in the genotype-phenotype map can induce complex and subtle behaviour in the dynamics that lead to enhanced evolutionary robustness and can be fruitfully described in terms of an elementary algorithmic "language". [ABSTRACT FROM AUTHOR]

This paper investigates the robustness of Run Transferable Libraries(RTLs) on scaled problems. RTLs provide GP with a library of functions which replace the usual primitive functions provided when approaching a problem. The RTL evolves from run to run using feedback based on function usage, and has been shown to outperform GP by an order of magnitude on a variety of scalable problems. RTLs can, however, also be applied across a domain of related problems, as well as across a range of scaled instances of a single problem. To do this successfully, it will need to balance a range of functions. We introduce a problem that can deceive the system into converging to a sub-optimal set of functions, and demonstrate that this is a consequence of the greediness of the library update algorithm. We demonstrate that a much simpler, truly evolutionary, update strategy doesn't suffer from this problem, and exhibits far better optimization properties than the original strategy. [ABSTRACT FROM AUTHOR]

We describe an approach to understanding evolved programs for a real world object detection problem, that of finding orthodontic landmarks in cranio-facial X-Rays. The approach involves modifying the fitness function to encourage the evolution of small programs, limiting the function set to a minimal number of operators and limiting the number of terminals (features). When this was done for two landmarks, an easy one and a difficult one, the evolved programs implemented a linear function of the features. Analysis of these linear functions revealed that underlying regularities were being captured and that successful evolutionary runs usually terminated with the best programs implementing one of a small number of underlying algorithms. Analysis of these algorithms revealed that they are a realistic solution to the object detection problem, given the features and operators available. [ABSTRACT FROM AUTHOR]

Genetic Programming (GP) has been shown to be a good method of predicting functions that solve inverse problems. In this context, a solution given by GP generally consists of a sole predictor. In contrast, Stack-based GP systems manipulate structures containing several predictors, which can be considered as teams of predictors. Work in Machine Learning reports that combining predictors gives good results in terms of both quality and robustness. In this paper, we use Stack-based GP to study different cooperations between predictors. First, preliminary tests and parameter tuning are performed on two GP benchmarks. Then, the system is applied to a real-world inverse problem. A comparative study with standard methods has shown limits and advantages of teams prediction, leading to encourage the use of combinations taking into account the response quality of each team member. [ABSTRACT FROM AUTHOR]

The commonly adopted fitness which evaluates the performance of individuals in co-evolutionary systems is the relative fitness. The relative fitness measure is a dynamic assessment subject to co-evolving population(s). Researchers apparently pay little attention to the use of absolute fitness functions in studying co-evolutionary algorithms. The first aim of this work is to define both the relative fitness and the absolute fitness for co-evolving systems. Another aim is to demonstrate the usage of the absolute and relative fitness through two case studies. One is for the Iterated Prisoners' Dilemma. Another case is for solving the Basic Alternating-Offers Bargaining Problem, for which a co-evolutionary system has been developed by means of Genetic Programming. Experiments using the relative fitness function have discovered co-adapted strategies that converge to nearly game-theoretic solutions. This finding suggests that the relative fitness essentially drives co-evolution to perfect equilibrium. On the other hand, the absolute fitness measuring co-evolving populations monitors the development of co-adaptation. Having analyzed the micro-behavior of the players' strategies based on their absolute fitness, we can explain how co-evolving populations stabilize at the perfect equilibrium. [ABSTRACT FROM AUTHOR]

In this paper, we present a general-purpose, systematic algorithm, consisting of a genetic programming module and a k-nearest neighbor classifier, to automatically create multiple artificial features (i.e., features that are computer-crafted and may not have a known physical meaning) directly from EEG signals, in a process that reveals patterns predictive of epileptic seizures. The algorithm was evaluated in three patients, with prediction defined over a horizon that varies between 1 and 5 minutes before unequivocal electrographic onset of seizure. For one patient, a perfect classification was achieved. For the other two patients, high classification accuracy was reached, predicting three seizures out of four for one, and eleven seizures out of fifteen for the other. For the latter, also, only one normal (non-seizure) signal was misclassified. These results compare favorably with other prediction approaches for patients from the same population. [ABSTRACT FROM AUTHOR]

A novel Grammatical Genetic Algorithm, the meta-Grammar Genetic Algorithm (mGGA) is presented. The mGGA borrows a grammatical representation and the ideas of modularity and reuse from Genetic Programming, and in particular an evolvable grammar representation from Grammatical Evolution by Grammatical Evolution. We demonstrate its application to a number of benchmark problems where significant performance gains are achieved when compared to static grammars. [ABSTRACT FROM AUTHOR]

This paper presents an algorithm for induction of ensembles of decision trees, also referred to as decision forests. In order to achieve high expressiveness the trees induced are multivariate, with various, possibly user-defined tests in their internal nodes. Strongly typed genetic programming is utilized to evolve structure of the tests. Special attention is given to the problem of diversity of the forest constructed. An approach is proposed, which explicitly encourages the induction algorithm to produce a different tree each run, which represents an alternative description of the data. It is shown that forests constructed this way have significantly reduced classification error even for small forest size, compared to other ensemble methods. Classification accuracy is also compared to other recent methods on several real-world datasets. [ABSTRACT FROM AUTHOR]

Particle Swarm Optimisers (PSOs) search using a set of interacting particles flying over the fitness landscape. These are typically controlled by forces that encourage each particle to fly back both towards the best point sampled by it and towards the swarm's best. Here we explore the possibility of evolving optimal force generating equations to control the particles in a PSO using genetic programming. [ABSTRACT FROM AUTHOR]

Computer games and simulations are commonly used as a basis for analysing and developing battlefield strategies. Such strategies are usually programmed explicitly, but it is also possible to generate them automatically via the use of evolutionary programming techniques. We focus in particular on the use of genetic programming to evolve strategies for a single defender facing multiple simultaneous attacks. By expressing the problem domain in the form of a ‘Space Invaders' game, we show that it is possible to evolve winning strategies for an increasingly complex sequence of scenarios. [ABSTRACT FROM AUTHOR]

This paper proposes a new architecture for tree-based genetic programming to evolve schema directly. It uses fixed length hs-expressions to represent program trees, keeps schema information in an instruction matrix, and extracts individuals from it. In order to manipulate the instruction matrix and the hs-expression, new genetic operators and new matrix functions are developed. The experimental results verify that its results are better than those of the canonical genetic programming on the problems tested in this paper. Keywords: IMGP, hs-expression, instruction matrix, schema evolution, Genetic Programming. [ABSTRACT FROM AUTHOR]

In real-time rendering, objects are represented using polygons or triangles. Triangles are easy to render and graphics hardware is highly optimized for rendering of triangles. Initially, the shading computations were carried out by dedicated hardwired algorithms for each vertex and then interpolated by the rasterizer. Todays graphics hardware contains vertex and pixel shaders which can be reprogrammed by the user. Vertex and pixel shaders allow almost arbitrary computations per vertex respectively per pixel. We have developed a system to evolve such programs. The system runs on a variety of graphics hardware due to the use of NVIDIA's high level Cg shader language. Fitness of the shaders is determined by user interaction. Both fixed length and variable length genomes are supported. The system is highly customizable. Each individual consists of a series of meta commands. The resulting Cg program is translated into the low level commands which are required for the particular graphics hardware. [ABSTRACT FROM AUTHOR]

In this paper the implementation of Genetic Programming (GP) to optimise a controller structure for a supply ship is assessed. GP is used to evolve control strategies for manoeuvring the ship. The optimised controllers are evaluated through computer simulations and real manoeuvrability tests in a water basin laboratory. In order to deal with the issue of generation of numerical constants, two kinds of GP algorithms are implemented. The first one chooses the constants necessary to create the control structure by random generation. The second algorithm includes a Genetic Algorithm (GA) for the optimisation of such constants. The results obtained illustrate the benefits of using GP to optimise propulsion and navigation controllers for ships. [ABSTRACT FROM AUTHOR]

Repeating code sequences are found in both artificial and natural genomes as an emergent phenomenon. These patterns are of interest in researching both how evolution reuses code segments to create superior individuals and whether building blocks are used in the formation of repeated sequences. We describe a GP representation using a special type of crossover that is more conducive to the formation of repeated sequences than traditional GP. We then establish that the repeated sequence phenomenon in the implementation displays traits of building blocks by establishing associated regularity of genotype and phenotype elements. [ABSTRACT FROM AUTHOR]

In this paper a new method is presented to solve a series of multiclass object classification problems using Genetic Programming (GP). All component two-class subproblems of the multiclass problem are solved in a single run, using a multi-objective fitness function. Probabilistic methods are used, with each evolved program required to solve only one subproblem. Programs gain a fitness related to their rank at the subproblem that they solve best. The new method is compared with two other GP based methods on four multiclass object classification problems of varying difficulty. The new method outperforms the other methods significantly in terms of both test classification accuracy and training time at the best validation performance in almost all experiments. [ABSTRACT FROM AUTHOR]

This paper introduces the Tree-String problem for genetic programming and related search and optimisation methods. To improve the understanding of optimisation and search methods, we aim to capture the complex dynamic created by the interdependencies of solution structure and content. Thus, we created an artificial domain that is amenable for analysis, yet representative of a wide-range of real-world applications. The Tree-String problem provides several benefits, including: the direct control of both structure and content objectives, the production of a rich and representative search space, the ability to create tunably difficult and random instances and the flexibility for specialisation. [ABSTRACT FROM AUTHOR]

In this paper we focus on machine-learning issues solved with Genetic Programming (GP). Excessive code growth or bloat often happens in GP, greatly slowing down the evolution process. In Pol03, Poli proposed the Tarpeian Control method to reduce bloat, but possible side-effects of this method on the generalization accuracy of GP hypotheses remained to be tested. In particular, since Tarpeian Control puts a brake on code growth, it could behave as a kind of Occam's razor, promoting shorter hypotheses more able to extend their knowledge to cases apart from any learning steps. To answer this question, we experiment Tarpeian Control with symbolic regression. The results are contrasted, showing that it can either increase or reduce the generalization power of GP hypotheses, depending on the problem at hand. Experiments also confirm the decrease in size of programs. We conclude that Tarpeian Control might be useful if carefully tuned to the problem at hand. [ABSTRACT FROM AUTHOR]

We extend our analysis of repetitive patterns found in genetic programming genomes to tree based GP. As in linear GP, repetitive patterns are present in large numbers. Size fair crossover limits bloat in automatic programming, preventing the evolution of recurring motifs. We examine these complex properties in detail: e.g. using depth v. size Catalan binary tree shape plots, subgraph and subtree matching, information entropy, syntactic and semantic fitness correlations and diffuse introns. We relate this emergent phenomenon to considerations about building blocks in GP and how GP works. [ABSTRACT FROM AUTHOR]

This paper explores distance measures based on genetic operators for genetic programming using tree structures. The consistency between genetic operators and distance measures is a crucial point for analytical measures of problem difficulty, such as fitness distance correlation, and for measures of population diversity, such as entropy or variance. The contribution of this paper is the exploration of possible definitions and approximations of operator-based edit distance measures. In particular, we focus on the subtree crossover operator. An empirical study is presented to illustrate the features of an operator-based distance. This paper makes progress toward improved algorithmic analysis by using appropriate measures of distance and similarity. [ABSTRACT FROM AUTHOR]

Genetic Parallel Programming (GPP) is a novel Genetic Programming paradigm. GPP Logic Circuit Synthesizer (GPPLCS) is a combinational logic circuit learning system based on GPP. The GPPLCS comprises a Multi-Logic-Unit Processor (MLP) which is a hardware processor built on a Field Programmable Gate Array (FPGA). The MLP is designed to speed up the evaluation of genetic parallel programs that represent combinational logic circuits. Four combinational logic circuit problems are presented to show the performance of the hardware-assisted GPPLCS. Experimental results show that the hardware MLP speeds up evolutions over 10 times. For difficult problems such as the 6-bit priority selector and the 6-bit comparator, the speedup ratio can be up to 22. [ABSTRACT FROM AUTHOR]

In this work we propose an approach of incorporating learning probabilistic context-sensitive grammar (LPCSG) in genetic programming (GP) employed for evolution and adaptation of locomotion gaits of simulated snake-like robot (Snakebot). In our approach LPCSG is derived from the originally defined context-free grammar, which usually expresses the syntax of genetic programs in canonical GP. During the especially introduced steered mutation the probabilities of applying each of particular production rules with multiple right-hand side alternatives in LPCSG depend on the context, and these probabilities are learned from the aggregated reward values obtained from the evolved best-of-generation Snakebots. Empirically obtained results verify that employing LPCSG contributes to the improvement of computational effort of both (i) the evolution of the fastest possible locomotion gaits for various fitness conditions and (ii) adaptation of these locomotion gaits to challenging environment and degraded mechanical abilities of Snakebot. In all of the cases considered in this study, the locomotion gaits, evolved and adapted employing GP with LPCSG feature higher velocity and are obtained faster than with canonical GP. Keywords: Snakebot, adaptation, locomotion, genetic programming, grammar. [ABSTRACT FROM AUTHOR]

This paper describes the first attempt to introduce evolutionarily designed players into the international Robocode league, a simulation-based game wherein robotic tanks fight to destruction in a closed arena. Using genetic programming to evolve tank strategies for this highly active forum, we were able to rank third out of twenty-seven players in the category of HaikuBots. Our GPBot was the only entry not written by a human. [ABSTRACT FROM AUTHOR]

We apply genetic programming to the evolution of strategies for playing the game of backgammon. Pitted in a 1000-game tournament against a standard benchmark player—Pubeval—our best evolved program wins 58% of the games, the highest verifiable result to date. Moreover, several other evolved programs attain win percentages not far behind the champion, evidencing the repeatability of our approach. [ABSTRACT FROM AUTHOR]

We apply genetic programming to the evolution of strategies for playing chess endgames. Our evolved programs are able to draw or win against an expert human-based strategy, and draw against CRAFTY—a world-class chess program, which finished second in the 2004 Computer Chess Championship. [ABSTRACT FROM AUTHOR]

We empirically investigate the use of dual duplication/truncation operators both as mutation operators and as generic local search operators, in combination with genetic search in a tree adjoining grammar guided genetic programming system (TAG3P). The results show that, on the problems tried, duplication/truncation works well as a mutation operator but not reliably when the complexity of the problem was scaled up. When using these dual operators as a generic local search operator, however, it helped TAG3P not only to solve the problems reliably but also cope well with scalability in problem complexity. Moreover, it managed to solve problems with very small population sizes. [ABSTRACT FROM AUTHOR]

Wireless sensor networks (WSNs) are medium scale manifestations of a paintable or amorphous computing paradigm. WSNs are becoming increasingly important as they attain greater deployment. New techniques for evolutionary computing (EC) are needed to address these new computing models. This paper describes a novel effort to develop a variation of traditional parallel evolutionary computing models to enable their use in the wireless sensor network. The ability to compute evolutionary algorithms within the WSN has innumerable advantages including intelligent-sensing, resource-optimized communication strategies, intelligent-routing protocol design, novelty detection, etc. In this paper we develop a parallel evolutionary algorithm suitable for use in a WSN. We then describe the adaptations required to develop practicable implementations to effectively operate in resource constrained environments such as WSNs. Several adaptations including a novel representation scheme, an approximate fitness computation method and a sufficient statistics based data reduction technique. These adaptations lead to the development of a GP implementation that is usable on the low-power, small footprint architectures typical to wireless sensor motes. We demonstrate the utility of our formulations and validate the proposed ideas using the algorithm to compute symbolic regression problems. [ABSTRACT FROM AUTHOR]

A protein is a linear chain of amino acids that folds into a unique functional structure, called its native state. In this state, proteins show repeated substructures like alpha helices and beta sheets. This suggests that native structures may be captured by the formalism known as Lindenmayer systems (L-systems). In this paper an evolutionary approach is used as the inference procedure for folded structures on simple lattice models. The algorithm searches the space of L-systems which are then executed to obtain the phenotype, thus our approach is close to Grammatical Evolution. The problem is to find a set of rewriting rules that represents a target native structure on the lattice model. The proposed approach has produced promising results for short sequences. Thus the foundations are set for a novel encoding based on L-systems for evolutionary approaches to both the Protein Structure Prediction and Inverse Folding Problems. [ABSTRACT FROM AUTHOR]

Cartesian Genetic Programming is a graph based representation that has many benefits over traditional tree based methods, including bloat free evolution and faster evolution through neutral search. Here, an integer based version of the representation is applied to a traditional problem in the field: evolving an obstacle avoiding robot controller. The technique is used to rapidly evolve controllers that work in a complex environment and with a challenging robot design. The generalisation of the robot controllers in different environments is also demonstrated. A novel fitness function based on chemical gradients is presented as a means of improving evolvability in such tasks. [ABSTRACT FROM AUTHOR]

This paper proposes the association of two approaches of GP which improve efficiency and reduce bloat. The first approach is to use a multi-population version of GP and the second one is to employ populations that can change size dynamically and adaptively. The latter approach consists in deleting or adding individuals in the population as a function of the current fitness and two other parameters. We test this approach on three well-known problems in GP, artificial ant, even parity 5 and one instance of the symbolic regression. We find that the combination of these two methods improves the quality of the individuals in the populations while keeping their size as small as possible and decreases the amount of resources required. [ABSTRACT FROM AUTHOR]

In this work a new approach, named Bayesian Automatic Programming (BAP), to inducing programs is presented. BAP integrates the power of grammar evolution and probabilistic models to evolve programs. We explore the use of BAP in two domains: a regression problem and the artificial ant problem. Its results are compared with traditional Genetic Programming (GP). The experimental results found encourage further investigation, especially to explore BAP in other domains and to improve the proposed approach to incorporating new mechanisms. [ABSTRACT FROM AUTHOR]

The three dozen or so known instances of human-competitive designs of antennas, mechanical systems, circuits, and controllers produced by genetic programming suggest the question of whether genetic programming can be extended to the design of complex structures from other fields. This paper describes how genetic programming can be used to automatically create a complete design for an optical lens system "from scratch"—without starting from a pre-existing good design and without pre-specifying the number of lenses, the layout of lenses, or the numerical parameters of the lenses. More particularly, genetic programming created an optical system that infringed a previous patent (the Konig patent) and improved upon another previous patent (the Tackaberry-Muller patent). The genetically evolved design is an example of a human-competitive result produced by genetic programming in the field of optical design. [ABSTRACT FROM AUTHOR]

Classical methods for Inductive Concept Learning (ICL) rely mostly on using specific search strategies, such as hill climbing and inverse resolution. These strategies have a great exploitation power, but run the risk of being incapable of escaping from local optima. An alternative approach to ICL is represented by Evolutionary Algorithms (EAs). EAs have a great exploration power, thus they have the capability of escaping from local optima, but their exploitation power is rather poor. These observations suggest that the two approaches are applicable to partly complementary classes of learning problems. More important, they indicate that a system incorporating features from both approaches could benefit from the complementary qualities of the approaches. In this paper we experimentally validate this statement. To this end, we incorporate different search strategies in a framework based on EAs for ICL. Results of experiments show that incorporating standard search strategies helps the EA in achieving better results. [ABSTRACT FROM AUTHOR]

Genetic Programming has been slow at realizing other programming paradigms than conventional, deterministic, sequential von-Neumann type algorithms. In this contribution we discuss a new method of execution of programs introduced recently: Algorithmic Chemistries. Therein, register machine instructions are executed in a non-deterministic order, following a probability distribution. Program behavior is thus highly dependent on frequency of instructions and connectivity between registers. Here we demonstrate the performance of GP on evolving solutions to a parity problem in a system of this type. [ABSTRACT FROM AUTHOR]