1 GAs and Feature Weighting Rebecca Fiebrink MUMT 611 31 March 2005.

Slides:



Advertisements
Similar presentations
Using Parallel Genetic Algorithm in a Predictive Job Scheduling
Advertisements

Tuesday, May 14 Genetic Algorithms Handouts: Lecture Notes Question: when should there be an additional review session?
Biologically Inspired AI (mostly GAs). Some Examples of Biologically Inspired Computation Neural networks Evolutionary computation (e.g., genetic algorithms)
EvoNet Flying Circus Introduction to Evolutionary Computation Brought to you by (insert your name) The EvoNet Training Committee The EvoNet Flying Circus.
Institute of Intelligent Power Electronics – IPE Page1 Introduction to Basics of Genetic Algorithms Docent Xiao-Zhi Gao Department of Electrical Engineering.
Monte Carlo Methods and the Genetic Algorithm Definitions and Considerations John E. Nawn MAT 5900 March 17 th, 2011.
1 Wendy Williams Metaheuristic Algorithms Genetic Algorithms: A Tutorial “Genetic Algorithms are good at taking large, potentially huge search spaces and.
Genetic Algorithms1 COMP305. Part II. Genetic Algorithms.
EvoNet Flying Circus Introduction to Evolutionary Computation Brought to you by (insert your name) The EvoNet Training Committee The EvoNet Flying Circus.
COMP305. Part II. Genetic Algorithms. Genetic Algorithms.
Introduction to Genetic Algorithms Yonatan Shichel.
COMP305. Part II. Genetic Algorithms. Genetic Algorithms.
COMP305. Part II. Genetic Algorithms. Genetic Algorithms.
Intro to AI Genetic Algorithm Ruth Bergman Fall 2002.
Selecting Informative Genes with Parallel Genetic Algorithms Deodatta Bhoite Prashant Jain.
Genetic Algorithms Nehaya Tayseer 1.Introduction What is a Genetic algorithm? A search technique used in computer science to find approximate solutions.
Intro to AI Genetic Algorithm Ruth Bergman Fall 2004.
Genetic Algorithm What is a genetic algorithm? “Genetic Algorithms are defined as global optimization procedures that use an analogy of genetic evolution.
Genetic Algorithms: A Tutorial
Genetic Algorithm.
A Genetic Algorithms Approach to Feature Subset Selection Problem by Hasan Doğu TAŞKIRAN CS 550 – Machine Learning Workshop Department of Computer Engineering.
Efficient Model Selection for Support Vector Machines
Genetic Algorithms CS121 Spring 2009 Richard Frankel Stanford University 1.
Cristian Urs and Ben Riveira. Introduction The article we chose focuses on improving the performance of Genetic Algorithms by: Use of predictive models.
Soft Computing Lecture 18 Foundations of genetic algorithms (GA). Using of GA.
SOFT COMPUTING (Optimization Techniques using GA) Dr. N.Uma Maheswari Professor/CSE PSNA CET.
Optimization in Engineering Design Georgia Institute of Technology Systems Realization Laboratory Mixed Integer Problems Most optimization algorithms deal.
Lecture 8: 24/5/1435 Genetic Algorithms Lecturer/ Kawther Abas 363CS – Artificial Intelligence.
Zorica Stanimirović Faculty of Mathematics, University of Belgrade
Genetic Algorithms Michael J. Watts
Applying Genetic Algorithm to the Knapsack Problem Qi Su ECE 539 Spring 2001 Course Project.
GENETIC ALGORITHMS FOR THE UNSUPERVISED CLASSIFICATION OF SATELLITE IMAGES Ankush Khandelwal( ) Vaibhav Kedia( )
An Introduction to Genetic Algorithms Lecture 2 November, 2010 Ivan Garibay
Genetic Algorithms Genetic Algorithms – What are they? And how they are inspired from evolution. Operators and Definitions in Genetic Algorithms paradigm.
Computational Complexity Jang, HaYoung BioIntelligence Lab.
Genetic Algorithms Introduction Advanced. Simple Genetic Algorithms: Introduction What is it? In a Nutshell References The Pseudo Code Illustrations Applications.
1 Machine Learning: Lecture 12 Genetic Algorithms (Based on Chapter 9 of Mitchell, T., Machine Learning, 1997)
Chapter 4.1 Beyond “Classic” Search. What were the pieces necessary for “classic” search.
1 “Genetic Algorithms are good at taking large, potentially huge search spaces and navigating them, looking for optimal combinations of things, solutions.
Soft Computing A Gentle introduction Richard P. Simpson.
Genetic Algorithms Siddhartha K. Shakya School of Computing. The Robert Gordon University Aberdeen, UK
GENETIC ALGORITHM A biologically inspired model of intelligence and the principles of biological evolution are applied to find solutions to difficult problems.
Derivative Free Optimization G.Anuradha. Contents Genetic Algorithm Simulated Annealing Random search method Downhill simplex method.
EE459 I ntroduction to Artificial I ntelligence Genetic Algorithms Kasin Prakobwaitayakit Department of Electrical Engineering Chiangmai University.
Evolution Programs (insert catchy subtitle here).
Chapter 9 Genetic Algorithms.  Based upon biological evolution  Generate successor hypothesis based upon repeated mutations  Acts as a randomized parallel.
Introduction to Genetic Algorithms. Genetic Algorithms We’ve covered enough material that we can write programs that use genetic algorithms! –More advanced.
Genetic Algorithms Genetic algorithms provide an approach to learning that is based loosely on simulated evolution. Hypotheses are often described by bit.
Chapter 12 FUSION OF FUZZY SYSTEM AND GENETIC ALGORITHMS Chi-Yuan Yeh.
EE749 I ntroduction to Artificial I ntelligence Genetic Algorithms The Simple GA.
Parallel Genetic Algorithms By Larry Hale and Trevor McCasland.
Feature Selection and Weighting using Genetic Algorithm for Off-line Character Recognition Systems Faten Hussein Presented by The University of British.
D Nagesh Kumar, IIScOptimization Methods: M8L5 1 Advanced Topics in Optimization Evolutionary Algorithms for Optimization and Search.
An Introduction to Genetic Algorithms Lecture 2 November, 2010 Ivan Garibay
Genetic Algorithms. Underlying Concept  Charles Darwin outlined the principle of natural selection.  Natural Selection is the process by which evolution.
Genetic Algorithm Dr. Md. Al-amin Bhuiyan Professor, Dept. of CSE Jahangirnagar University.
Artificial Intelligence By Mr. Ejaz CIIT Sahiwal Evolutionary Computation.
CAP6938 Neuroevolution and Artificial Embryogeny Evolutionary Comptation Dr. Kenneth Stanley January 23, 2006.
1 Comparative Study of two Genetic Algorithms Based Task Allocation Models in Distributed Computing System Oğuzhan TAŞ 2005.
Genetic Algorithms. Solution Search in Problem Space.
EVOLUTIONARY SYSTEMS AND GENETIC ALGORITHMS NAME: AKSHITKUMAR PATEL STUDENT ID: GRAD POSITION PAPER.
Genetic Algorithms An Evolutionary Approach to Problem Solving.
Genetic Algorithm(GA)
Genetic Algorithm. Outline Motivation Genetic algorithms An illustrative example Hypothesis space search.
 Presented By: Abdul Aziz Ghazi  Roll No:  Presented to: Sir Harris.
Genetic Algorithms.
Genetic Algorithms: A Tutorial
Traveling Salesman Problem by Genetic Algorithm
Genetic Algorithms: A Tutorial
Presentation transcript:

1 GAs and Feature Weighting Rebecca Fiebrink MUMT March 2005

2 of 25 Outline Genetic algorithms –History of GAs –How GAs work Simple model Variations Feature selection and weighting –Feature selection –Feature weighting Using GAs for feature weighting –Siedlecki and Sklansky –Punch –The ACE project

3 Part 1: Genetic Algorithms

4 of 25 History of GAs 1859: Charles Darwin, Origin of Species 1950’s – 60’s: Computers simulate evolution 1960’s – 70’s: John Holland invents genetic algorithms –Adaptation in Natural and Artificial Systems, book published 1975 –Context of “adaptive systems,” not just “optimization” 1980’s – Present: Further exploration, widespread adoptation –Kenneth De Jong –Goldberg: Genetic Algorithms in Search, Optimization, and Machine Learning (“classic” textbook published 1989) Related areas: evolutionary programming, genetic programming (Carnegie Mellon GA online FAQ, Cantu-Paz 2000)

5 of 25 How GAs work Problem is a search space –Variable parameters  dimensions –Problem has minima and maxima within this space –Minima and maxima are hard to find Potential solutions describe points in this space –It is “easy” to calculate “goodness” of any one solution and compare it to other solutions Maintain a set of potential solutions –Guess a set of initial solutions –Combine these solutions with each other –Keep “good” solutions and discard “bad” solutions –Repeat combination and selection process for some time

6 of 25 A simple GA Start Terminate? Stop Select Parents Produce Offspring Mutate Offspring Evaluate Population

7 of 25 GA terminology Population: The set of all current potential solutions Deme: A subset of the population that interbreeds (might be the whole population) Chromosome: A solution structure (often binary); contains one or more genes Gene: A feature or parameter Allele: A specific value for a gene Phenotype: A member of the population, a real-valued solution vector Generation: A complete cycle of evaluation, selection, crossover, and mutation in a deme Locus: A particular position (bit) on the chromosome string Crossover: The process of combining two chromosomes; simplest method is single-point crossover where two chromosomes swap parts on either side of a random locus Mutation: A random change to a phenotype (i.e., changing an allele)

8 of 25 Crossover Illustrated Parent 1 Child 1 Parent 2 Child 2

9 of 25 GA Variations Variables: –Population size –Selection method –Crossover method –Mutation rate –How to encode a solution in a chromosome No single choice is best for all problems (Wolpert & Macready 1997, cited in Cantu- Paz 2000).

10 of 25 More variations: Parallel GAs Categories: –Single-population Master/Slave –Multiple population –Fine-grained –Hybrids Master Slaves (Cantu-Paz 2000)

11 of 25 Applications of GAs NP problems (e.g., Traveling Salesman) Airfoil design Noise control Fluid dynamics Circuit partitioning Image processing Liquid crystals Water networks Music (Miettinen et al. 1999, Coley 1999)

12 Part 2: Feature Selection and Weighting

13 of 25 Features What are they? –A classifier’s “handle” to the problem –Numerical or binary values representing independent or dependent attributes of each instance to be classified What are they in music? –Spectral centroid –Number of instruments –Composer –Presence of Banjo –Beat histogram - …

14 of 25 Feature Selection – Why? “Curse of dimensionality”: –The size of the training set must grow exponentially with the dimensionality of the feature space The set of best features to use may vary according to classification scheme, goal, or data set We need a way to select only a good subset of all possible features –May or may not be interested in “best” subset

15 of 25 Feature Selection – How? Common approaches –Dimensionality reduction (principle component analysis or factor analysis) –Experimentation: Choose a subset, train a classifier with it, and evaluate its performance. Example: –A piece might have vector (values) –Try selections, … ? (1=yes, 0=no) –Which works best? Setbacks in experimentation: –For n potential features, there are 2 n possible subsets: a huge search space! – Evaluating each classifier takes time –Choose vectors using sequential search, branch, and bound, or GAs

16 of 25 Feature Weighting Among a group of selected (useful) features, some may be more useful than others Weight the features for optimal classifier performance Experimental approach: Similar to feature selection –Example: Say is optimal selection for feature set Try weights like,, … ? Practical and theoretical constraints of feature selection are magnified

17 Part 3: Using GAs for Feature Weighting

18 of 25 GAs and Feature Weighting A chromosome is a vector of weights –Each gene corresponds to a feature weight E.g., for selection E.g., for weighting The fitness of a chromosome is a measure of its performance training and validating an actual classifier

19 of 25 Siedlecki and Sklansky 1989: A note on genetic algorithms for large-scale feature selection Choose feature subset for Knn classifier Propose using GAs when feature set > 20 Binary chromosomes (0/1 for each feature) Goal: seek the smallest/least costly subset of features for which classifier performance above a certain level

20 of 25 Results Compared GAs with exhaustive search, branch and bound, and (p,q)-search on sample problem –Branch and bound and (p, q)-search did not come close enough (within 1% error) to finding a set that performed optimally –Exhaustive search used 2 24 evaluations –GA found optimal or near-optimal solutions with evaluations, a huge savings over both exhaustive search and branch and bound GA exhibits slightly more than linear increase of time complexity with added features GA also outperforms other methods on a real problem with 30 features

21 of 25 Punch et al. 1993: Further research on feature selection and classification using genetic algorithms Approach: Use GAs for feature weighting for Knn classifier: classifier “dimension warping” Each chromosome is a vector of real-valued weights –Mapped exponentially to range [.01, 10) or 0 Also experimented with “hidden features” representing combination (multiplication) of 2 other features

22 of 25 Results Feature weighting can outperform simple selection, especially on large data set with noise –Best performance when feature weighting follows binary selection 14 days to compute results Parallel version: nearly linear speedup when strings passed to other processors for fitness evaluation

23 of 25 Recent work Minaei-Bidgoli, Kortemeyer, and Punch, 2004: Optimizing classification ensembles via a genetic algorithm for a web-based educational system –Incorporate GA feature weighting in a multiple-classifier system (vote system) –Results show over 10% improvement in accuracy of the classifier ensemble when GA optimization is used

24 of 25 MIR Application ACE project –Accept arbitrary type and number of features, arbitrary taxonomy, training and testing data, and a target classification time –Classify data using best means available; e.g., make use of multiple classifiers, feature weighting –Parallel GAs for feature weighting can improve solution time and quality

25 of 25 Conclusions GAs are powerful and interesting tools, but they are complex and not always well-understood Feature selection/weighting involves selecting from a huge space of possibilities, but it is a necessary task GAs are useful for tackling the feature weighting problem Faster machines, parallel computing, and better understanding of GA behavior can all benefit the feature weighting problem This is relevant to MIR: we want to do good and efficient classification!