Ahmed Kattan

Using semantics in the selection mechanism in Genetic Programming: A simple method for promoting semantic diversity

Research on semantics in Genetic Programming (GP) has increased over the last number of years. Results in this area clearly indicate that its use in GP considerably increases performance. Many of these semantic-based approaches rely on a trial-and-error method that attempts to find offspring that are semantically different from their parents over a number of trials using the crossover operator (crossover-semantics based - CSB). This, in consequence, has a major drawback: these methods could evaluate thousands of nodes, resulting in paying a high computational cost, while attempting to improve performance by promoting semantic diversity. In this work, we propose a simple and computationally inexpensive method, named semantics in selection, that eliminates the computational cost observed in CSB approaches. We tested this approach in 14 GP problems, including continuous- and discrete-valued fitness functions, and compared it against a traditional GP and a CSB approach. Our results are equivalent, and in some cases, superior than those found by the CSB approach, without the necessity of using a “brute force” mechanism.

Classification of localized muscle fatigue with genetic programming on sEMG during isometric contraction

Genetic Programming is used to generate a solution that can classify localized muscle fatigue from filtered and rectified surface electromyography (sEMG). The GP has two classification phases, the GP training phase and a GP testing phase. In the training phase, the program evolved with multiple components. One component analyzes statistical features extracted from sEMG to chop the signal into blocks and label them using a fuzzy classifier into three classes: Non-Fatigue, Transition-to-Fatigue and Fatigue. The blocks are then projected onto a two-dimensional Euclidean space via two further (evolved) program components. K-means clustering is then applied to group similar data blocks. Each cluster is then labeled according to its dominant members. The programs that achieve good classification are evolved. In the testing phase, it tests the signal using the evolved components, however without the use of a fuzzy classifier. As the results show the evolved program achieves good classification and it can be used on any unseen isometric sEMG signals to classify fatigue without requiring any further evolution. The GP was able to classify the signal into a meaningful sequence of Non-Fatigue→Transition-to-Fatigue→Fatigue. By identifying a Transition-to Fatigue state the GP can give a prediction of an oncoming fatigue. The genetic classifier gave promising results 83.17% correct classification on average of all signals in the test set, especially considering that the GP is classifying muscle fatigue for ten different individuals.

Evolving radial basis function networks via GP for estimating fitness values using surrogate models

In real-world problems with candidate solutions that are very expensive to evaluate, Surrogate Models (SMs) mimic the behaviour of the simulation model as closely as possible while being computationally cheaper to evaluate. Due to their nature, SMs can be seen as heuristics that can help to estimate the fitness of a candidate solution without having to evaluate it. In this paper, we propose a new SM based on Genetic Programming (GP) and Radial Basis Function Networks (RBFN), called GP-RBFN Surrogate. More specifically, we use GP to evolve both: the structure of a RBF and its parameters. The SM evolved by our algorithm is tested in one of the most studied NP-complete problem (MAX-SAT) and its performance is compared against RBFN Surrogate, GAs, Random Search and (1+1) ES. The results obtained by performing extensive empirical experiments indicate that our proposed approach outperforms the other four methods in terms of finding better solutions without the need of evaluating a large portion of candidate solutions.

Universal intelligent data compression systems: A review

Researchers have classically addressed the problem of universal compression using two approaches. The first approach has been to develop adaptive compression algorithms, where the system changes its behaviour during the compression to fit the encoding situation of the given data. The second approach has been to use the composition of multiple compression algorithms. Recently, however, a third approach has been adopted by researchers in order to develop compression systems: the application of computational intelligence paradigms. This has shown remarkable results in the data compression domain improving the decision making process and outperforming conventional systems of data compression. This paper reviews some of the previous attempts to address the universal compression problem within conventional and computational intelligence techniques.

Evolutionary synthesis of lossless compression algorithms with GP-zip3

Here we propose GP-zip3, a system which uses Genetic Programming to find optimal ways to combine standard compression algorithms for the purpose of compressing files and archives. GP-zip3 evolves programs with multiple components. One component analyses statistical features extracted from the raw data to be compressed (seen as a sequence of 8-bit integers) to divide the data into blocks. These blocks are then projected onto a two-dimensional Euclidean space via two further (evolved) program components. K-means clustering is applied to group similar data blocks. Each cluster is then labelled with the optimal compression algorithm for its member blocks. Once a program that achieves good compression is evolved, it can be used on unseen data without the requirement for any further evolution. GP-zip3 is similar to its predecessor, GP-zip2. Both systems outperform a variety of standard compression algorithms and are faster than other evolutionary compression techniques. However, GP-zip2 was still substantially slower than off-the-shelf algorithms. GP-zip3 alleviates this problem by using a novel fitness evaluation strategy. More specifically, GP-zip3 evolves and then uses decision trees to predict the performance of GP individuals without requiring them to be used to compress the training data. As shown in a variety of experiments, this speeds up evolution in GP-zip3 considerably over GP-zip2 while achieving similar compression results, thereby significantly broadening the scope of application of the approach.

Evolving optimal agendas for package deal negotiation

This paper presents a hyper GA system to evolve optimal agendas for package deal negotiation. The proposed system uses a Surrogate Model based on Radial Basis Function Networks (RBFNs) to speed up the evolution. The negotiation scenario is as follows. There are two negotiators/agents (a and b) and m issues/items available for negotiation. But from these m issues, the agents must choose g issues and negotiate on them. The g issues thus chosen form the agenda. The agenda is important because the outcome of negotiation depends on it. Furthermore, a and b will, in general, get different utilities/profits from different agendas. Thus, for competitive negotiation (i.e., negotiation where each agent wants to maximize its own utility), each agent wants to choose an agenda that maximizes its own profit. However, the problem of determining an agents optimal agenda is complex, as it requires combinatorial search. To overcome this problem, we present a hyper GA method that uses a Surrogate Model based on Radial Basis Function Networks (RBFNs) to find an agents optimal agenda. The performance of the proposed method is evaluated experimentally. The results of these experiments demonstrate that the surrogate assisted algorithm, on average, performs better than standard GA and random search.

GP made faster with semantic surrogate modelling

Genetic Programming (GP) is known to be expensive in cases where the fitness evaluation is computationally demanding, i.e., object detection, programmatic compression, image processing applications. The paper introduces a method that reduces the amount of fitness evaluations that are required to obtain good solutions. We consider the supervised learning setting, where a training set of input vectors are collectively mapped to a vector of outputs, and then a loss function is used to map the vector of outputs to a scalar fitness value. Saving of fitness evaluations is achieved through the use of two components. The first component is surrogate model that predicts trees output for a particular input vector xi based on the similarity between xi and other input vectors in the training set for which the candidate solution has been already evaluated with. The second component, is a simple linear equation to control the size of a sub-training set that is used to train GP trees. This linear equation allows the size of the sub-training set to dynamically increase or decrease based on the status of the search. The proposed method referred to as SSGP. Empirical results in 17 different problems, from three different categories, demonstrate that SSGP is able to obtain solutions of similar quality with those obtained using several benchmark GP systems, but with a much smaller computation time. The simplicity of the proposed method and the ease of its implementation is one of the most appealing aspects of its future utility.

Evolution of human-competitive lossless compression algorithms with GP-zip2

We propose GP-zip2, a new approach to lossless data compression based on Genetic Programming (GP). GP is used to optimally combine well-known lossless compression algorithms to maximise data compression. GP-zip2 evolves programs with multiple components. One component analyses statistical features extracted by sequentially scanning the data to be compressed and divides the data into blocks. These blocks are projected onto a two-dimensional Euclidean space via two further (evolved) program components. K-means clustering is then applied to group similar data blocks. Each cluster is labelled with the optimal compression algorithm for its member blocks. After evolution, evolved programs can be used to compress unseen data. The compression algorithms available to GP-zip2 are: Arithmetic coding, Lempel-Ziv-Welch, Unbounded Prediction by Partial Matching, Run Length Encoding, and Bzip2. Experimentation shows that the results produced by GP-zip2 are human-competitive, being typically superior to well-established human-designed compression algorithms in terms of the compression ratios achieved in heterogeneous archive files.

Locality in Continuous Fitness-Valued Cases and Genetic Programming Difficulty

It is commonly accepted that a mapping is local if it preserves neighbourhood. In Evolutionary Computation, locality is generally described as the property that neighbouring genotypes correspond to neighbouring phenotypes. Locality has been classified in one of two categories: high and low locality. It is said that a representation has high locality if most genotypic neighbours correspond to phenotypic neighbours. The opposite is true for a representation that has low locality. It is argued that a representation with high locality performs better in evolutionary search compared to a representation that has low locality. In this work, we explore, for the first time, a study on Genetic Programming (GP) locality in continuous fitnessvalued cases. For this, we extended the original definition of locality (first defined and used in Genetic Algorithms using bitstrings) from genotype-phenotype mapping to the genotype-fitness mapping. Then, we defined three possible variants of locality in GP regarding neighbourhood. The experimental tests presented here use a set of symbolic regression problems, two different encoding and two different mutation operators. We show how locality can be studied in this type of scenarios (continuous fitness-valued cases) and that locality can successfully been used as a performance prediction tool.

Speaker Verification on Unbalanced Data with Genetic Programming

Automatic Speaker Verification (ASV) is a highly unbalanced binary classification problem, in which any given speaker must be verified against everyone else. We apply Genetic programming (GP) to this problem with the aim of both prediction and inference. We examine the generalisation of evolved programs using a variety of fitness functions and data sampling techniques found in the literature. A significant difference between train and test performance, which can indicate overfitting, is found in the evolutionary runs of all to-be-verified speakers. Nevertheless, in all speakers, the best test performance attained is always superior than just merely predicting the majority class. We examine which features are used in good-generalising individuals. The findings can inform future applications of GP or other machine learning techniques to ASV about the suitability of feature-extraction techniques.