Modeling XCS in class imbalances: Population sizing and parameter settings

Jul 13, 20070 likes564 views

This paper analyzes the scalability of the population size required in XCS to maintain niches that are infrequently activated. Facetwise models have been developed to predict the effect of the imbalance ratio—ratio between the number of instances of the majority class and the minority class that are sampled to XCS—on population initialization, and on the creation and deletion of classifiers of the minority class. While theoretical models show that, ideally, XCS scales linearly with the imbalance ratio, XCS with standard configuration scales exponentially. The causes that are potentially responsible for this deviation from the ideal scalability are also investigated. Specifically, the inheritance procedure of classifiers’ parameters, mutation, and subsumption are analyzed, and improvements in XCS’s mechanisms are proposed to effectively and efficiently handle imbalanced problems. Once the recommendations are incorporated to XCS, empirical results show that the population size in XCS indeed scales linearly with the imbalance ratio.

Modeling XCS in class imbalances: Population sizing and parameter settings

This paper presents a learning methodology based on a substructural classification model to solve decomposable classification problems. The proposed method consists of three important components: (1) a structural model that represents salient interactions between attributes for a given data, (2) a surrogate model which provides a functional approximation of the output as a function of attributes, and (3) a classification model which predicts the class for new inputs. The structural model is used to infer the functional form of the surrogate and its coefficients are estimated using linear regression methods. The classification model uses a maximally-accurate, least-complex surrogate to predict the output for given inputs. The structural model that yields an optimal classification model is searched using an iterative greedy search heuristic. Results show that the proposed method successfully detects the interacting variables in hierarchical problems, group them in linkages groups, and build maximally accurate classification models. The initial results on non-trivial hierarchical test problems indicate that the proposed method holds promise and have also shed light on several improvements to enhance the capabilities of the proposed method.

IWLCS'2007: Substructural Surrogates for Learning Decomposable Classification...Albert Orriols-Puig

The document presents a methodology for building surrogates to aid in the classification of decomposable problems. It consists of three layers: 1) a structural model layer that extracts dependencies between attributes, 2) a surrogate model layer that induces a functional form and estimates coefficients based on the structural model, and 3) a classification model layer that uses the surrogate to predict classes for new examples. The methodology is tested on hierarchical problems with lower level blocks of either 2 or 3 bits. Results show the surrogates capture the salient interactions between variables, outperforming decision trees and support vector machines which include irrelevant attributes in their models.

Modeling selection pressure in XCS for proportionate and tournament selectionkknsastry

In this paper, we derive models of the selection pressure in XCS for proportionate (roulette wheel) selection and tournament selection. We show that these models can explain the empirical results that have been previously presented in the literature. We validate the models on simple problems showing that, (i) when the model assumptions hold, the theory perfectly matches the empirical evidence; (ii) when the model assumptions do not hold, the theory can still provide qualitative explanations of the experimental results.

Translational bioinformatics at VHIR: Understanding molecular damage in Fabry...Vall d'Hebron Institute of Research (VHIR)

This document summarizes research on developing a computational model to identify pathological mutations in Fabry disease. The researchers built a neural network model using 7 properties of mutations including sequence, structure, and multiple sequence alignment data. They trained and tested the model on a dataset of 313 pathological and 59 neutral mutations, achieving a sensitivity of 0.85 and specificity of 0.92. The model outperformed general tools like Polyphen-2 in predicting mutations for Fabry disease. Future work will focus on expanding the approach to additional genes and disease phenotypes.

Mm@iknow09Michael Kohlegger

This document discusses understanding maturity models through a structured content analysis of existing maturity models. It presents the methodology used, which was a structured content analysis of 16 maturity models from different domains. The analysis sought to better understand the concept of maturity modelling and inform future developments, particularly in the area of knowledge management. Key findings from the analysis include implications for defining concepts like what maturity means, what the maturing subject is, how models are used, and who uses them.

Virscidian Poster Asms2010 Final Version LetterMark Bayliss

New challenges monolixday2011blaudez

The document summarizes activities of POPIX and DDMoRe related to population modelling and the clinical trial simulator. Key points include: 1) POPIX develops new population modelling methods in fields like pharmacology while partnering with DDMoRe on PK/PD modelling. 2) DDMoRe aims to establish modelling standards and shares disease models through its modelling library and framework. 3) POPIX works on flexible statistical models, Bayesian estimation, errors in design/covariates, hidden Markov models, and stochastic differential equation models. 4) The clinical trial simulator can simulate trials using various PKPD models and recruitment/compliance models, and integrate workflows for simulation and analysis.

Identification of Causal Variables for Building Energy Fault Detection by Sem...Keigo Yoshida

Leadership style and perceived benefits of electronic data interchange for re...rsd kol abundjani

This dissertation examines the relationship between leadership style and perceptions of benefits from electronic data interchange (EDI) in the retail industry. Survey data was collected from 90 retail executives using validated scales to measure leadership style and perceived EDI benefits. Linear regression analysis found that executives with more transformational leadership styles perceived greater usefulness of EDI, while those with more transactional styles perceived EDI as easier to use. The study aims to provide insights into how leadership impacts views of information technology benefits in supply chain management.

Stephen Friend Institute of Development, Aging and Cancer 2011-11-28Sage Base

Stephen Friend presented on using data intensive science and bionetworks to build better models of disease. He discussed how current disease models often oversimplify pathways and indications, failing to capture disease complexity. Integrating large datasets using computational models hosted in "compute spaces" could generate more accurate disease maps. However, barriers include a lack of data sharing, standardized tools and incentives. Sage Bionetworks aims to establish a commons for collaborative model building to accelerate disease understanding and new therapies.

Damage-based models for flexural response on RC elementsMaria Grazia Santoro

Object recognition with cortex like mechanisms pami-07dingggthu

This document summarizes a new framework for robust object recognition inspired by the visual cortex. It describes a hierarchical system that builds an increasingly complex and invariant feature representation through alternating template matching and maximum pooling operations. The approach demonstrates strong performance on single object recognition, multiclass categorization, and scene understanding tasks. Given its biological constraints, it performs surprisingly well and competes with state-of-the-art systems while learning from few examples. The success of this cortex-like model provides plausibility for feedforward models of object recognition in the visual cortex.

信号検出理論の解説 (Signal detection theory, a primer)Masatoshi Yoshida

FA no miss ○ CR ● The document describes the process of signal detection theory (SDT). It involves: 1) Collecting training data samples from known signals and noises. 2) Setting criteria for optimal discrimination between signals and noises. 3) Classifying new test data by comparing the sensor values to the discrimination criteria. The goal is to accurately detect signals while minimizing incorrect classifications of noise as signals.

Extreme Spatio-Temporal Data AnalysisJoel Saltz

The document discusses extreme spatio-temporal data analysis in biomedical informatics. It summarizes contributions in computer science methods for analyzing large datasets from sensors and in biomedical fields to mine insights. The talk outlines analyzing pathology images, tumor subtyping in brain tumors, whole slide image analysis in clinical practice, and tissue flow analysis using high performance computing.

Genetic Programming based Image SegmentationTarundeep Dhot

Komogortsev Qualitative And Quantitative Scoring And Evaluation Of The Eye Mo...Kalle

This paper presents a set of qualitative and quantitative scores designed to assess performance of any eye movement classification algorithm. The scores are designed to provide a foundation for the eye tracking researchers to communicate about the performance validity of various eye movement classification algorithms. The paper concentrates on the five algorithms in particular: Velocity Threshold Identification (I-VT), Dispersion Threshold Identification (I-DT), Minimum Spanning Tree Identification (MST), Hidden Markov Model Identification (IHMM) and Kalman Filter Identification (I-KF). The paper presents an evaluation of the classification performance of each algorithm in the case when values of the input parameters are varied. Advantages provided by the new scores are discussed. Discussion on what is the "best" classification algorithm is provided for several applications. General recommendations for the selection of the input parameters for each algorithm are provided.

Labmatrixkimjjones02

The document describes a research informatics software system called Labmatrix. It allows users to track clinical and research data in a hierarchical and relational manner. The system facilitates secure collaborative data entry and easy ad-hoc data retrieval. It can capture detailed patient and specimen data, generate barcodes, track specimen storage locations and lineage. Users can update multiple records simultaneously and generate sample records using predefined workflows. Data can be imported from various sources and normalized. The system includes graphical tools for exploring and querying the data without requiring SQL or programming skills.

Stephen Friend Molecular Imaging Program at Stanford (MIPS) 2011-08-15Sage Base

The document discusses using "data intensive science" and building better disease maps through comprehensive monitoring of disease and molecular traits in large populations. It describes constructing co-expression networks from gene expression measures across hundreds of samples to identify modules of genes that interact. Preliminary probabilistic models have been built using these networks to directly identify genes that are causal for disease.

Zander eng scd_finalJustyna Zander

This document summarizes Justyna Zander-Nowicka's doctoral thesis defense on December 19th, 2008 regarding her research on model-based testing of embedded real-time systems in the automotive domain. The thesis proposed a model-based testing approach called MiLEST that uses signal features for automatic test data generation and evaluation. The approach aims to systematically generate functional test cases from models to test embedded systems starting from early development phases.

Optimizing Requirements Decisions with KEYSgregoryg

The document discusses optimizing requirements models using a new algorithm called KEYS. KEYS beats other standard algorithms like simulated annealing and state of the art methods. KEYS allows requirements models to be optimized 50,000 times faster, in almost real-time. KEYS works by first identifying "keys" or important variables that influence the overall model, and then optimizing settings for those key variables. The document provides details on how KEYS finds keys using a technique called BORE sampling. It also discusses using KEYS to optimize a requirements model called DDP used at JPL for designing deep space missions.

Sparse feature analysis for detection of clustered microcalcifications in mam...Wesley De Neve

This document analyzes the use of sparse feature analysis for detecting clustered microcalcifications in mammogram images. It compares different feature types, combinations of features, and dictionary construction techniques for sparse representation based classification (SRC) of mammogram images. The experimental results show that texture features like Laws' texture features (LAW) are more effective than shape/morphology features. SRC using LAW features alone or combined with local binary patterns (LBP) achieved high performance. Larger dictionaries containing more atoms resulted in higher discriminative power for the SRC-based detection system.

GMatasci_Talk_DomainSeparationForEfficientAdaptiveAL_IGARSS2011.pdfgrssieee

This document presents a method for combining domain separation and adaptive active learning. It aims to efficiently label samples from a new target domain using available labeled data from a related source domain. The method first uses domain separation to identify informative regions of the target domain distinct from the source. It then applies adaptive active learning via breaking ties to iteratively select the most uncertain target samples for labeling. A TrAdaBoost algorithm is used to differentially reweight source and target samples as more target data is incorporated, in order to adapt the classifier to the target domain. The method is tested on QuickBird imagery of Zurich from different seasons, demonstrating a dataset shift. Results show the combined approach can effectively label samples for the target domain using minimal supervision

Extracting Proximity for Brain Graph Voxel ClassificationJoshua Vogelstein

White-Box Software Testing using Evolutionary AlgorithmsYaser Sulaiman

Exploiting Semantic Structure for Mapping User-specified Form Terms to SNOMED...The Children's Hospital of Philadelphia

The document presents a method for automatically mapping terms from clinical encounter forms to concepts in SNOMED CT. It exploits the semantic structure of forms by analyzing the context and relationships between terms. A naive Bayes classifier is trained on semantic attributes derived from the form structure to determine the appropriate SNOMED CT semantic category for each term. Evaluation on 26 forms showed the hybrid approach of combining linguistic techniques and semantic structure outperformed a baseline, achieving a precision of 0.89 and recall of 0.76.

Stephen Friend Norwegian Academy of Science and Letters 2011-11-02Sage Base

The document discusses using data intensive science to build better models of disease. It argues that the current pharmaceutical model is broken because it lacks sufficient understanding of disease biology. Most drug candidates fail because existing disease models oversimplify complex conditions. The author proposes using large datasets and computational modeling to map molecular pathways and construct causal models of diseases. This could provide a more mechanistic understanding of diseases and their heterogeneity to identify true drivers and better predict treatment responses. Pilots are needed to determine if this approach can modify diseases rather than just treat symptoms by moving beyond lists of altered genes and proteins to causal networks.

Stephen Friend IBC Next Generation Sequencing & Genomic Medicine 2011-08-03Sage Base

This document discusses a drug development approach called Arch2POCM that moves from disease targets to clinical validation. It focuses on using data intensive science to build better disease maps by generating massive amounts of data from comprehensive population monitoring and integrating genotypic, gene expression, and trait data into probabilistic models. This allows the direct identification of causal genes for disease.

GECCO'2007: Modeling XCS in Class Imbalances: Population Size and Parameter S...Albert Orriols-Puig

1. The document describes a facetwise analysis of the XCS learning classifier system for class imbalances. 2. It analyzes the population initialization process, generation of rules for minority classes, time to extinction of such rules, and derives a population size bound. 3. The analysis considers problems with multiple classes, one sampled at a lower frequency (minority class), and derives probabilities of sampling instances from each class.

GECCO'2006: Bounding XCS’s Parameters for Unbalanced DatasetsAlbert Orriols-Puig

This document discusses bounding the parameters of the XCS learning algorithm for imbalanced datasets. It outlines analyzing XCS's performance on imbalanced data, the contribution of its components, and approaches to facilitate learning minority class regions. The document will describe XCS and the domain, present experimentation, discuss how XCS handles class imbalances, provide guidelines for parameter tuning, consider online adaptation, and draw conclusions.

IWLCS'05: The Class Imbalance Problem in Learning Classifier Systems: A Prel...Albert Orriols-Puig

More Related Content

What's hot (19)

Leadership style and perceived benefits of electronic data interchange for re...rsd kol abundjani

Stephen Friend Institute of Development, Aging and Cancer 2011-11-28Sage Base

Damage-based models for flexural response on RC elementsMaria Grazia Santoro

Object recognition with cortex like mechanisms pami-07dingggthu

信号検出理論の解説 (Signal detection theory, a primer)Masatoshi Yoshida

Extreme Spatio-Temporal Data AnalysisJoel Saltz

Genetic Programming based Image SegmentationTarundeep Dhot

Komogortsev Qualitative And Quantitative Scoring And Evaluation Of The Eye Mo...Kalle

Labmatrixkimjjones02

Stephen Friend Molecular Imaging Program at Stanford (MIPS) 2011-08-15Sage Base

Zander eng scd_finalJustyna Zander

Optimizing Requirements Decisions with KEYSgregoryg

Sparse feature analysis for detection of clustered microcalcifications in mam...Wesley De Neve

GMatasci_Talk_DomainSeparationForEfficientAdaptiveAL_IGARSS2011.pdfgrssieee

Extracting Proximity for Brain Graph Voxel ClassificationJoshua Vogelstein

White-Box Software Testing using Evolutionary AlgorithmsYaser Sulaiman

Exploiting Semantic Structure for Mapping User-specified Form Terms to SNOMED...The Children's Hospital of Philadelphia

Stephen Friend Norwegian Academy of Science and Letters 2011-11-02Sage Base

Stephen Friend IBC Next Generation Sequencing & Genomic Medicine 2011-08-03Sage Base

Leadership style and perceived benefits of electronic data interchange for re...rsd kol abundjani

Stephen Friend Institute of Development, Aging and Cancer 2011-11-28Sage Base

Damage-based models for flexural response on RC elementsMaria Grazia Santoro

Object recognition with cortex like mechanisms pami-07dingggthu

信号検出理論の解説 (Signal detection theory, a primer)Masatoshi Yoshida

Extreme Spatio-Temporal Data AnalysisJoel Saltz

Genetic Programming based Image SegmentationTarundeep Dhot

Komogortsev Qualitative And Quantitative Scoring And Evaluation Of The Eye Mo...Kalle

Labmatrixkimjjones02

Stephen Friend Molecular Imaging Program at Stanford (MIPS) 2011-08-15Sage Base

Zander eng scd_finalJustyna Zander

Optimizing Requirements Decisions with KEYSgregoryg

Sparse feature analysis for detection of clustered microcalcifications in mam...Wesley De Neve

GMatasci_Talk_DomainSeparationForEfficientAdaptiveAL_IGARSS2011.pdfgrssieee

Extracting Proximity for Brain Graph Voxel ClassificationJoshua Vogelstein

White-Box Software Testing using Evolutionary AlgorithmsYaser Sulaiman

Exploiting Semantic Structure for Mapping User-specified Form Terms to SNOMED...The Children's Hospital of Philadelphia

Stephen Friend Norwegian Academy of Science and Letters 2011-11-02Sage Base

Stephen Friend IBC Next Generation Sequencing & Genomic Medicine 2011-08-03Sage Base

Similar to Modeling XCS in class imbalances: Population sizing and parameter settings (20)

GECCO'2007: Modeling XCS in Class Imbalances: Population Size and Parameter S...Albert Orriols-Puig

GECCO'2006: Bounding XCS’s Parameters for Unbalanced DatasetsAlbert Orriols-Puig

IWLCS'05: The Class Imbalance Problem in Learning Classifier Systems: A Prel...Albert Orriols-Puig

Introducing LCS to Digital Design VerificationDaniele Loiacono

The document introduces the use of learning classifier systems (LCS) for digital design verification. It defines what a design and design verification are, and explains that design verification aims to discover bugs before manufacturing by maximizing test coverage. The document notes that design verification is challenging due to increasing design complexity and limited resources. It summarizes previous attempts using genetic algorithms, genetic programming, Bayesian networks and Markov models that achieved good results but had limitations. The document proposes using an LCS called XCS to overcome these limitations through adaptive learning and developing an accurate problem representation.

New Challenges in Learning Classifier Systems: Mining Rarities and Evolving F...Albert Orriols-Puig

The document discusses new challenges in learning classifier systems (LCS) when dealing with domains containing rare classes. It proposes using a design decomposition approach to analyze how LCS address rare classes. Specifically, it examines how the extended classifier system (XCS) handles rare classes. It identifies five critical elements of LCS that are important for detecting small niches associated with rare classes: 1) estimating classifier parameters correctly, 2) providing representatives of rare niches during initialization, 3) generating and growing representatives of rare niches, 4) adjusting the genetic algorithm application rate, and 5) ensuring representatives of rare niches dominate their niches. The document focuses on analyzing the first element of estimating classifier parameters for XCS when dealing with domains

CCIA'2008: Can Evolution Strategies Improve Learning Guidance in XCS? Design ...Albert Orriols-Puig

This document proposes using evolution strategies (ES) instead of genetic algorithms (GA) in the genetic algorithm component of the XCS learning classifier system. It designs an ES-based XCS with a modified classifier representation and ES-based genetic operators. Experiments on real-world datasets show the ES-based XCS outperforms GA-based XCS with selection and mutation alone, though there is no significant difference when crossover is added. Further research is suggested to determine when different search operators should be used.

633-600 Machine Learningbutest

This document contains information about a 633-600 Machine Learning course, including: 1. Contact information for the instructor, Yoonsuck Choe, and a link to the course webpage. 2. An overview of the course content including textbook information, grading criteria, academic integrity policies, and lecture notes. 3. A high-level introduction to machine learning concepts like how machines can learn, current applications, and the multidisciplinary roots of the field.

Evolutionary Testing of Stateful Systems: a Holistic Approachmatteomiraz

The document summarizes research on evolutionary testing of stateful systems. It discusses limitations of traditional search-based software testing approaches when applied to object-oriented systems, and proposes a new approach called Testful. Testful uses an evolutionary algorithm to generate test cases for a class under test and its dependencies. It aims to maximize both test coverage and compactness of tests. The paper describes improvements to Testful including local search, seeding, and fitness inheritance techniques to enhance efficiency. It then evaluates Testful's ability to achieve control flow and data flow coverage compared to other automated testing tools.

Learning Classifier Systems for Class Imbalance ProblemsXavier Llorà

The document discusses learning classifier systems (LCS) for addressing class imbalance problems in datasets. It aims to enhance the applicability of LCS to knowledge discovery from real-world datasets that often exhibit class imbalance, where one class is represented by significantly fewer examples than other classes. The author proposes adapting parameters of the XCS learning classifier system, such as learning rate and genetic algorithm threshold, based on estimated class imbalance ratios within classifiers' niches in order to minimize bias towards majority classes and better handle small disjuncts representing minority classes.

CEC'2005: Class Imbalance Problem in UCS Classifier System: Fitness AdaptationAlbert Orriols-Puig

The document discusses the class imbalance problem in the UCS (Unordered Classifier System) machine learning system. It describes UCS and how it was tested on datasets with varying levels of class imbalance. The results show that UCS's performance decreases as the class imbalance increases. The document then covers methods for UCS to deal with class imbalances, such as class-sensitive accuracy and weighted class-sensitive accuracy metrics.

HIS'2008: New Crossover Operator for Evolutionary Rule Discovery in XCSAlbert Orriols-Puig

This document proposes a new crossover operator called BLX crossover for use in XCS, an evolutionary learning classifier system. BLX crossover combines the innovation power of two-point crossover with local search by allowing the boundaries of classifier rules to move during crossover. Experiments on 12 real-world datasets show that BLX crossover enables XCS to more accurately fit complex decision boundaries compared to two-point crossover, and may prevent overfitting. The work demonstrates the importance of further research on genetic algorithm operators for evolutionary rule discovery.

Pattern classification via unsupervised learnersNick Palmer

I study classification problems in a standard learning framework, in which an unsupervised learner creates a discriminant function over each class and observations are labeled by the learner returning the highest value associated with that observation. I examine whether this approach gains significant advantage over traditional discriminant techniques. Probably Approximately Correct learning distributions over class labels under L1 distance or KL-divergence is shown to imply PAC classification in this framework. I determine bounds on the regret associated with the resulting classifier, taking into account the possibility of variable misclassification penalties, and demonstrate the advantage of estimating the a posteriori probability distributions over class labels in the setting of Optical Character Recognition. Unsupervised learners can be used to learn a class of probabilistic concepts (stochastic rules denoting the probability that an observation has a positive label in a 2-class setting). I demonstrate a situation where unsupervised learners can be used even when it is hard to learn distributions over class labels – in this case the discriminant functions do not estimate the class probability densities. If that isn't exciting enough, I then use a standard state-merging technique to PAC-learn a class of probabilistic automata. The results show that by learning the distribution over outputs under the weaker L1 distance rather than KL-divergence we are able to learn without knowledge of the expected length of an output. It is also shown that for a restricted class of these automata learning under L1 distance is equivalent to learning under KL-divergence.

Transfer Learning, Soft Distance-Based Bias, and the Hierarchical BOAMartin Pelikan

An automated technique has recently been proposed to transfer learning in the hierarchical Bayesian optimization algorithm (hBOA) based on distance-based statistics. The technique enables practitioners to improve hBOA efficiency by collecting statistics from probabilistic models obtained in previous hBOA runs and using the obtained statistics to bias future hBOA runs on similar problems. The purpose of this paper is threefold: (1) test the technique on several classes of NP-complete problems, including MAXSAT, spin glasses and minimum vertex cover; (2) demonstrate that the technique is effective even when previous runs were done on problems of different size; (3) provide empirical evidence that combining transfer learning with other efficiency enhancement techniques can often yield nearly multiplicative speedups.

Test Optimization Using Adaptive Random Testing TechniquesQuEST Global (erstwhile NeST Software)

In today’s competitive software development scenario, the customer demands a testing coverage which not only ensures the stated requirements but also the implied ones. This situation calls for an exhaustive testing which may not be always possible due to various reasons. Testing, due to its last position in SDLC, often gets crunched due to the cumulative schedule slippages. Hence Tester is faced with a challenge to make testing as efficient as possible within a short time span due to cost constraints. With selective testing an only option, test leads usually go for the age-old approach of Random Testing. Random testing does not ensure coverage in a scientific manner.

On the value of stochastic analysis for software engineeringCS, NcState

S10butest

1. The document discusses machine learning and provides an overview of key concepts like inductive reasoning, learning from examples, and the constituents of machine learning problems. 2. It explains that machine learning problems involve an example set, background concepts, background axioms, and potential errors in data. Common machine learning tasks are categorization and prediction. 3. The document also outlines the constituents of machine learning methods, including representation schemes, search methods, and approaches for selecting hypotheses when multiple solutions are produced.

S10butest

lecture_01.pptx - PowerPoint Presentationbutest

This document summarizes the key points from the first lecture of a course on pattern recognition. It introduces common applications of pattern recognition, related fields like machine learning and artificial intelligence. It emphasizes that understanding the properties of the data and incorporating prior knowledge are important for developing effective pattern recognition systems. The course will focus on decision theory, statistical learning methods, and evaluating algorithms rather than on specific feature extraction techniques.

HIS'2008: Genetic-based Synthetic Data Sets for the Analysis of Classifiers B...Albert Orriols-Puig

This document presents a genetic algorithm approach to generating synthetic data sets for analyzing classifier behavior. The genetic algorithm represents data set labelings as binary strings and uses genetic operators like crossover and mutation to evolve solutions that satisfy the desired complexity based on class boundary length. Experiments show the genetic algorithm can generate intermediate complexity data sets in early generations and produce similar accuracy rates across different classifier paradigms, while allowing control over the data set properties. Future work aims to improve efficiency and scalability, enable multiple criteria optimization, and develop benchmark problems with more realistic structure.

Exact Computation of the Expectation Curves of the Bit-Flip Mutation using La...jfrchicanog

The document presents research on calculating the expected fitness of solutions after applying a bit-flip mutation operator using landscape theory. It defines key concepts like landscape, elementary landscapes, and landscape decomposition. The main results are a simple formula to compute the expected fitness for elementary landscapes and a generalized formula for arbitrary functions by decomposing into elementary components. Potential applications to genetic algorithms, simulated annealing, and other metaheuristics are discussed. Future work ideas include expressions for higher moments and probabilities, new operators based on the results, and connections to runtime analysis.

GECCO'2007: Modeling XCS in Class Imbalances: Population Size and Parameter S...Albert Orriols-Puig

GECCO'2006: Bounding XCS’s Parameters for Unbalanced DatasetsAlbert Orriols-Puig

IWLCS'05: The Class Imbalance Problem in Learning Classifier Systems: A Prel...Albert Orriols-Puig

Introducing LCS to Digital Design VerificationDaniele Loiacono

New Challenges in Learning Classifier Systems: Mining Rarities and Evolving F...Albert Orriols-Puig

CCIA'2008: Can Evolution Strategies Improve Learning Guidance in XCS? Design ...Albert Orriols-Puig

633-600 Machine Learningbutest

Evolutionary Testing of Stateful Systems: a Holistic Approachmatteomiraz

Learning Classifier Systems for Class Imbalance ProblemsXavier Llorà

CEC'2005: Class Imbalance Problem in UCS Classifier System: Fitness AdaptationAlbert Orriols-Puig

HIS'2008: New Crossover Operator for Evolutionary Rule Discovery in XCSAlbert Orriols-Puig

Pattern classification via unsupervised learnersNick Palmer

Transfer Learning, Soft Distance-Based Bias, and the Hierarchical BOAMartin Pelikan

Test Optimization Using Adaptive Random Testing TechniquesQuEST Global (erstwhile NeST Software)

On the value of stochastic analysis for software engineeringCS, NcState

S10butest

lecture_01.pptx - PowerPoint Presentationbutest

HIS'2008: Genetic-based Synthetic Data Sets for the Analysis of Classifiers B...Albert Orriols-Puig

Exact Computation of the Expectation Curves of the Bit-Flip Mutation using La...jfrchicanog

More from kknsastry (20)

Genetic Algorithms and Genetic Programming for Multiscale Modelingkknsastry

Effective and efficient multiscale modeling is essential to advance both the science and synthesis in a wide array of fields such as physics, chemistry, materials science, biology, biotechnology and pharmacology. This study investigates the efficacy and potential of using genetic algorithms for multiscale materials modeling and addresses some of the challenges involved in designing competent algorithms that solve hard problems quickly, reliably and accurately.

Towards billion bit optimization via parallel estimation of distribution algo...kknsastry

The document describes research into using efficient estimation of distribution algorithms (EDAs) like the compact genetic algorithm (cGA) to solve optimization problems involving billions of variables. Key aspects discussed include the cGA's memory and computational efficiency through techniques like parallelization and vectorization. The researchers were able to solve a noisy OneMax problem involving over 33 million variables to optimality and a problem with 1.1 billion variables with relaxed convergence. The document argues this research is important because many real-world problems involving nanotechnology, biology, and information systems require solving optimization problems at massive scales.

Empirical Analysis of ideal recombination on random decomposable problemskknsastry

This paper analyzes the behavior of a selectorecombinative genetic algorithm (GA) with an ideal crossover on a class of random additively decomposable problems (rADPs). Specifically, additively decomposable problems of order k whose subsolution fitnesses are sampled from the standard uniform distribution U[0,1] are analyzed. The scalability of the selectorecombinative GA is investigated for 10,000 rADP instances. The validity of facetwise models in bounding the population size, run duration, and the number of function evaluations required to successfully solve the problems is also verified. Finally, rADP instances that are easiest and most difficult are also investigated.

Population sizing for entropy-based model buliding In genetic algorithms kknsastry

This paper presents a population-sizing model for the entropy-based model building in genetic algorithms. Specifically, the population size required for building an accurate model is investigated. The effect of the selection pressure on population sizing is also incorporated. The proposed model indicates that the population size required for building an accurate model scales as Θ(m log m), where m is the number of substructures and proportional to the problem size. Experiments are conducted to verify the derivations, and the results agree with the proposed model.

Let's get ready to rumble redux: Crossover versus mutation head to head on ex...kknsastry

This paper analyzes the relative advantages between crossover and mutation on a class of deterministic and stochastic additively separable problems with substructures of non-uniform salience. This study assumes that the recombination and mutation operators have the knowledge of the building blocks (BBs) and effectively exchange or search among competing BBs. Facetwise models of convergence time and population sizing have been used to determine the scalability of each algorithm. The analysis shows that for deterministic exponentially-scaled additively separable, problems, the BB-wise mutation is more efficient than crossover yielding a speedup of Θ(l logl), where l is the problem size. For the noisy exponentially-scaled problems, the outcome depends on whether scaling on noise is dominant. When scaling dominates, mutation is more efficient than crossover yielding a speedup of Θ(l logl). On the other hand, when noise dominates, crossover is more efficient than mutation yielding a speedup of Θ(l).

Let's get ready to rumble redux: Crossover versus mutation head to head on ex...kknsastry

This paper analyzes the relative advantages between crossover and mutation on a class of deterministic and stochastic additively separable problems with substructures of non-uniform salience. This study assumes that the recombination and mutation operators have the knowledge of the building blocks (BBs) and effectively exchange or search among competing BBs. Facetwise models of convergence time and population sizing have been used to determine the scalability of each algorithm. The analysis shows that for deterministic exponentially-scaled additively separable, problems, the BB-wise mutation is more efficient than crossover yielding a speedup of Θ(l logl), where l is the problem size. For the noisy exponentially-scaled problems, the outcome depends on whether scaling on noise is dominant. When scaling dominates, mutation is more efficient than crossover yielding a speedup of Θ(l logl). On the other hand, when noise dominates, crossover is more efficient than mutation yielding a speedup of Θ(l).

Automated alphabet reduction with evolutionary algorithms for protein structu...kknsastry

This paper focuses on automated procedures to reduce the dimensionality of protein structure prediction datasets by simplifying the way in which the primary sequence of a protein is represented. The potential benefits of this procedure are faster and easier learning process as well as the generation of more compact and human-readable classifiers. The dimensionality reduction procedure we propose consists on the reduction of the 20-letter amino acid (AA) alphabet, which is normally used to specify a protein sequence, into a lower cardinality alphabet. This reduction comes about by a clustering of AA types accordingly to their physical and chemical similarity. Our automated reduction procedure is guided by a fitness function based on the Mutual Information between the AA-based input attributes of the dataset and the protein structure feature that being predicted. To search for the optimal reduction, the Extended Compact Genetic Algorithm (ECGA) was used, and afterwards the results of this process were fed into (and validated by) BioHEL, a genetics-based machine learning technique. BioHEL used the reduced alphabet to induce rules for protein structure prediction features. BioHEL results are compared to two standard machine learning systems. Our results show that it is possible to reduce the size of the alphabet used for prediction from twenty to just three letters resulting in more compact, i.e. interpretable, rules. Also, a protein-wise accuracy performance measure suggests that the loss of accuracy accrued by this substantial alphabet reduction is not statistically significant when compared to the full alphabet.

Analyzing probabilistic models in hierarchical BOA on traps and spin glasseskknsastry

The hierarchical Bayesian optimization algorithm (hBOA) can solve nearly decomposable and hierarchical problems of bounded difficulty in a robust and scalable manner by building and sampling probabilistic models of promising solutions. This paper analyzes probabilistic models in hBOA on two common test problems: concatenated traps and 2D Ising spin glasses with periodic boundary conditions. We argue that although Bayesian networks with local structures can encode complex probability distributions, analyzing these models in hBOA is relatively straightforward and the results of such analyses may provide practitioners with useful information about their problems. The results show that the probabilistic models in hBOA closely correspond to the structure of the underlying optimization problem, the models do not change significantly in subsequent iterations of BOA, and creating adequate probabilistic models by hand is not straightforward even with complete knowledge of the optimization problem.

Modeling selection pressure in XCS for proportionate and tournament selectionkknsastry

Modeling XCS in class imbalances: Population size and parameter settingskknsastry

Fast and accurate reaction dynamics via multiobjective genetic algorithm opti...kknsastry

The document describes research into using multiobjective genetic algorithms to optimize semiempirical potentials for fast and accurate reaction dynamics simulations. The researchers developed a method to tune semiempirical parameters using a limited set of ab initio calculations to better describe excited state potential energy surfaces. They found that multiobjective optimization was able to find globally accurate potential energy surfaces more efficiently than weighted single-objective optimizations. Analysis of the optimized parameter sets showed they produced stable and physically reasonable results.

On Extended Compact Genetic Algorithmkknsastry

In this study we present a detailed analysis of the extended compact genetic algorithm (ECGA). Based on the analysis, empirical relations for population sizing and convergence time have been derived and are compared with the existing relations. We then apply ECGA to a non-azeotropic binary working fluid power cycle optimization problem. The optimal power cycle obtained improved the cycle efficiency by 2.5% over that existing cycles, thus illustrating the capabilities of ECGA in solving real-world problems.

Silicon Cluster Optimization Using Extended Compact Genetic Algorithmkknsastry

This paper presents an efficient cluster optimization algorithm. The proposed algorithm uses extended compact genetic algorithm (ECGA), one of the competent genetic algorithms (GAs) coupled with Nelder-Mead simplex local search. The lowest energy structures of silicon clusters with 4-11 atoms have been successfully predicted. The minimum population size and total number of function (potential energy of the cluster) evaluations required to converge to the global optimum with a reliability of 96% have been empirically determined and are O(n4.2) and O(n8.2) respectively. The results obtained indicate that the proposed algorithm is highly reliable in predicting globally optimal structures. However, certain efficiency techniques have to be employed for predicting structures of larger clusters to reduce the high computational cost due to function evaluation.

A Practical Schema Theorem for Genetic Algorithm Design and Tuningkknsastry

This paper develops the theory that can enable the design of genetic algorithms and choose the parameters such that the proportion of the best building blocks grow. A practical schema theorem has been used for this purpose and its ramification for the choice of selection operator and parameterization of the algorithm is explored. In particular stochastic universal selection, tournament selection, and truncation selection schemes are employed to verify the results. Results agree with the schema theorem and indicate that it must be obeyed in order to ascertain sustained growth of good building blocks. The analysis suggests that schema theorem alone is insufficient to guarantee the success of a selectorecombinative genetic algorithm.

On the Supply of Building Blockskknsastry

This study addresses the issue of building-block supply in the initial population. Facetwise models for supply of a single building block as well as for supply of all schemata in a partition have been developed. An estimate for the population size required to ensure the presence of all raw building blocks has been derived using these facetwise models. The facetwise models and the population-sizing estimate are verified with computational results.

Don't Evaluate, Inheritkknsastry

This paper studies fitness inheritance as an efficiency enhancement technique for genetic and evolutionary algorithms. Convergence and population-sizing models are derived and compared with experimental results. These models are optimized for greatest speed-up and the optimal inheritance proportion to obtain such a speed-up is derived. Results on OneMax problems show that when the inheritance effects are considered in the population-sizing model, the number of function evaluations are reduced by 20% with the use of fitness inheritance. Results indicate that for a fixed population size, the number of function evaluations can be reduced by 70% using a simple fitness inheritance technique.

Efficient Cluster Optimization Using A Hybrid Extended Compact Genetic Algori...kknsastry

A recent study Sastry and Xiao (2001) proposed a highly reliable cluster optimization algorithm which employed extended compact genetic algorithm (ECGA) along with Nelder-Mead simplex search. This study utilizes an efficiency enhancement technique for the ECGA based cluster optimizer to reduce the population size and the number of function evaluation requirements, yet retaining the high reliability of predicting the lowest energy structure. Seeding of initial population with lowest energy structures of smaller cluster has been employed as the efficiency enhancement technique. Empirical results indicate that the population size and total number of function evaluations scale up with the cluster size are reduced from O(n4.2) and O(n8.2) to O(n0.83) and O(n2.45) respectively.

Modeling Tournament Selection with Replacement Using Apparent Added Noisekknsastry

This paper analyzes the effects of tournament selection with replacement on the convergence time and population sizing for selectorecombinative genetic algorithms. This paper empirically demonstrates that the run duration remains the same and is not affected whether the tournament selection is performed with or without replacement. However, the population size required is more if tournament selection is performed with replacement rather than without replacement to attain the same level of accuracy. An approximate population-sizing model is derived based on apparent added noise for the case of tournament selection with replacement. The proposed model is verified with experimental results.

Evaluation Relaxation as an Efficiency-Enhancement Technique: Handling Varian...kknsastry

This study develops a decision-making strategy for deciding between fitness functions with differing bias values. Simple, yet practical facetwise models are derived to aid the decision-making process. The decision making strategy is designed to provide maximum speed-up and thereby enhance the efficiency of GA search processes. Results indicate that bias can be handled temporally and that significant speed-up values can be obtained.

Analysis of Mixing in Genetic Algorithms: A Surveykknsastry

Ensuring building-block (BB) mixing is critical to the success of genetic and evolutionary algorithms. There has been a growing interest in analyzing and understanding BB mixing and it is necessary to organize and categorize representative literature. This paper presents an exhaustive survey of studies on one or more aspects of mixing. In doing so, a classification of the literature based on the role of recombination operators assumed by those studies is developed. Such a classification not only highlights the significant results and unifies existing work, but also provides a foundation for future research in understanding mixing in genetic algorithms.