Using Genetic Programming to Learn Probability Distributions as Mutation Operators with Evolutionary Programming Libin Hong, John Woodward, Ender Ozcan,

Slides:



Advertisements
Similar presentations
Applications of one-class classification
Advertisements

Ali Husseinzadeh Kashan Spring 2010
Biologically Inspired Computation Really finishing off EC.
On the Genetic Evolution of a Perfect Tic-Tac-Toe Strategy
UNIVERSITY OF JYVÄSKYLÄ Building NeuroSearch – Intelligent Evolutionary Search Algorithm For Peer-to-Peer Environment Master’s Thesis by Joni Töyrylä
EvoNet Flying Circus Introduction to Evolutionary Computation Brought to you by (insert your name) The EvoNet Training Committee The EvoNet Flying Circus.
DAN SHMIDT ITAY BITTAN Advanced Topics in Evolutionary Algorithms.
Estimation of Distribution Algorithms Let’s review what have done in EC so far: We have studied EP and found that each individual searched via Gaussian.
COMP305. Part II. Genetic Algorithms. Genetic Algorithms.
Evolutionary Design By: Dianna Fox and Dan Morris.
Slightly beyond Turing’s computability for studying Genetic Programming Olivier Teytaud, Tao, Inria, Lri, UMR CNRS 8623, Univ. Paris-Sud, Pascal, Digiteo.
Fast Evolutionary Optimisation Temi avanzati di Intelligenza Artificiale - Lecture 6 Prof. Vincenzo Cutello Department of Mathematics and Computer Science.
Intro to AI Genetic Algorithm Ruth Bergman Fall 2002.
Tutorial 1 Temi avanzati di Intelligenza Artificiale - Lecture 3 Prof. Vincenzo Cutello Department of Mathematics and Computer Science University of Catania.
Selecting Informative Genes with Parallel Genetic Algorithms Deodatta Bhoite Prashant Jain.
Experimental Evaluation
CS 447 Advanced Topics in Artificial Intelligence Fall 2002.
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 Programming. Agenda What is Genetic Programming? Background/History. Why Genetic Programming? How Genetic Principles are Applied. Examples of.
Transformation of Input Space using Statistical Moments: EA-Based Approach Ahmed Kattan: Um Al Qura University, Saudi Arabia Michael Kampouridis: University.
Prepared by Barış GÖKÇE 1.  Search Methods  Evolutionary Algorithms (EA)  Characteristics of EAs  Genetic Programming (GP)  Evolutionary Programming.
Genetic Programming on Program Traces as an Inference Engine for Probabilistic Languages Vita Batishcheva, Alexey Potapov
In this paper we propose 10,000,000 algorithms: The (semi-)automatic design of (heuristic) algorithms Introducing John Woodward Edmund Burke, Gabriela.
The Automatic Generation of MutationOperators.pptx for Genetic Algorithms [Workshop on Evolutionary Computation for the Automated Design of Algorithms.
Efficient Model Selection for Support Vector Machines
Adapting Convergent Scheduling Using Machine Learning Diego Puppin*, Mark Stephenson †, Una-May O’Reilly †, Martin Martin †, and Saman Amarasinghe † *
Integrating Neural Network and Genetic Algorithm to Solve Function Approximation Combined with Optimization Problem Term presentation for CSC7333 Machine.
Cristian Urs and Ben Riveira. Introduction The article we chose focuses on improving the performance of Genetic Algorithms by: Use of predictive models.
Hierarchical Distributed Genetic Algorithm for Image Segmentation Hanchuan Peng, Fuhui Long*, Zheru Chi, and Wanshi Siu {fhlong, phc,
Lecture 8: 24/5/1435 Genetic Algorithms Lecturer/ Kawther Abas 363CS – Artificial Intelligence.
Boltzmann Machine (BM) (§6.4) Hopfield model + hidden nodes + simulated annealing BM Architecture –a set of visible nodes: nodes can be accessed from outside.
Introduction to Evolutionary Algorithms Session 4 Jim Smith University of the West of England, UK May/June 2012.
Optimization of multi-pass turning operations using ant colony system Authors: K. Vijayakumar, G. Prabhaharan, P. Asokan, R. Saravanan 2003 Presented by:
TEMPLATE DESIGN © Zhiyao Duan 1,2, Lie Lu 1, and Changshui Zhang 2 1. Microsoft Research Asia (MSRA), Beijing, China.2.
GENETIC ALGORITHMS FOR THE UNSUPERVISED CLASSIFICATION OF SATELLITE IMAGES Ankush Khandelwal( ) Vaibhav Kedia( )
Evolving Virtual Creatures & Evolving 3D Morphology and Behavior by Competition Papers by Karl Sims Presented by Sarah Waziruddin.
Genetic Algorithms Introduction Advanced. Simple Genetic Algorithms: Introduction What is it? In a Nutshell References The Pseudo Code Illustrations Applications.
Jun-Won Suh Intelligent Electronic Systems Human and Systems Engineering Department of Electrical and Computer Engineering Speaker Verification System.
GENETIC ALGORITHM A biologically inspired model of intelligence and the principles of biological evolution are applied to find solutions to difficult problems.
Evolutionary Programming
MAE 552 Heuristic Optimization Instructor: John Eddy Lecture #12 2/20/02 Evolutionary Algorithms.
CITS7212: Computational Intelligence An Overview of Core CI Technologies Lyndon While.
Improving Support Vector Machine through Parameter Optimized Rujiang Bai, Junhua Liao Shandong University of Technology Library Zibo , China { brj,
Contribution of second order evolution to evolutionary algorithms Virginie LEFORT July 11 th.
1 Autonomic Computer Systems Evolutionary Computation Pascal Paysan.
Application of the GA-PSO with the Fuzzy controller to the robot soccer Department of Electrical Engineering, Southern Taiwan University, Tainan, R.O.C.
Principles in the Evolutionary Design of Digital Circuits J. F. Miller, D. Job, and V. K. Vassilev Genetic Programming and Evolvable Machines.
1 ParadisEO-MOEO for a Bi-objective Flow-Shop Scheduling Problem May 2007 E.-G. Talbi and the ParadisEO team
Genetic Algorithm Dr. Md. Al-amin Bhuiyan Professor, Dept. of CSE Jahangirnagar University.
Selection and Recombination Temi avanzati di Intelligenza Artificiale - Lecture 4 Prof. Vincenzo Cutello Department of Mathematics and Computer Science.
Agenda  INTRODUCTION  GENETIC ALGORITHMS  GENETIC ALGORITHMS FOR EXPLORING QUERY SPACE  SYSTEM ARCHITECTURE  THE EFFECT OF DIFFERENT MUTATION RATES.
Metaheuristics for the New Millennium Bruce L. Golden RH Smith School of Business University of Maryland by Presented at the University of Iowa, March.
Deep Belief Network Training Same greedy layer-wise approach First train lowest RBM (h 0 – h 1 ) using RBM update algorithm (note h 0 is x) Freeze weights.
Evolutionary Programming A.E. Eiben and J.E. Smith, Introduction to Evolutionary Computing Chapter 5.
A.E. Eiben and J.E. Smith, Introduction to Evolutionary Computing Evolutionary Programming.
Advanced AI – Session 7 Genetic Algorithm By: H.Nematzadeh.
Evolutionary Computation Evolving Neural Network Topologies.
Introduction to genetic algorithm
Evolutionary Programming
USING MICROBIAL GENETIC ALGORITHM TO SOLVE CARD SPLITTING PROBLEM.
Genetic Improvement CHORDS GROUP Stirling University
Sample vs Population comparing mean and standard deviations
Artificial Intelligence Chapter 4. Machine Evolution
Dr. Kenneth Stanley February 6, 2006
Aiman H. El-Maleh Sadiq M. Sait Syed Z. Shazli
Artificial Intelligence Chapter 4. Machine Evolution
EE368 Soft Computing Genetic Algorithms.
Evolutionary Programming
Beyond Classical Search
Presentation transcript:

Using Genetic Programming to Learn Probability Distributions as Mutation Operators with Evolutionary Programming Libin Hong, John Woodward, Ender Ozcan, Jingpeng Li The university of Nottingham The University of Stirling

Summary of Abstract In Nutshell 1.Evolutionary programing optimizes functions by evolving a population of real-valued vectors (genotype). 2.Variation has been provided (manually) by probability distributions (Gaussian, Cauchy, Levy). 3.We are automatically generating probability distributions (using genetic programming). 4.Not from scratch, but from already well known distributions (Gaussian, Cauchy, Levy). We are “genetically improving probability distributions”. 5.We are evolving mutation operators for a problem class (a probability distributions over functions). Genotype is (1.3,...,4.5,…,8.7) Before mutation Genotype is (1.2,...,4.4,…,8.6) After mutation

Outline of Talk 1.Concept (Automatic vs. Manual Design) 2.Benchmarking Function Instances, Function Classes. 3.Function Optimization by Evolutionary Programming (Gaussian/Cauchy mutation). 4.Genetic Programming to design distributions. 5.Experimental Results.

Optimization & Benchmark Functions A set of 23 benchmark functions is typically used in the literature. Minimization We use the first 10 but as problem classes.

Function Class 1 (of 10) 1.Machine learning needs to generalize. 2.We generalize to function classes. 3.y = x ^ 2 (a function) 4.y = a x ^ 2 (parameterised function) 5.y = a x ^ 2, a ~[1,2] (function class) 6.We do this for all 10 (23) functions. 7.Function classes are naturally occurring in domains (not forced for the sake of this paper). 8.The probability distribution we evolve fits the problem class.

Probability Distributions for Evolutionary Programming 1. Function optimization by Evolutionary Programming. 2. A population of real-valued vectors is varied “mutated” (perturbed) to generate new vectors which undergo evolutionary competition. 3. Mutation is typically provided by Gaussian and Cauchy probability distributions. 4. Can probability distributions be automatically generated which outperform the human nominated probability distributions?

Gaussian and Cauchy Distributions 1.A Gaussian distribution is used to model noise. 2.A Cauchy distribution is generated by one Gaussian divided by another Gaussian. 3.Cauchy (large jumps) good at start of search. 4.Gaussian (smaller jumps) good at end of search. CAUCHY GAUSSIAN

(Fast) Evolutionary Programming 1.EP mutates with a Gaussian. 2.Fast EP mutates with a Cauchy. 3.A generalization is mutate with a distribution D (generated with genetic programming) Heart of algorithm is mutation SO LETS AUTOMATICALLY DESIGN

The 2 Dimensional Version of f8 Which is the best mutation operator, Gaussian or Cauchy distribution? Lets design a distribution automatically!

Meta and Base Learning At the base level we are learning about a specific function. At the meta level we are learning about the problem class. We are just doing “generate and test” at a higher level What is being passed with each blue arrow? Conventional EP EP Function to optimize Probability Distribution Generator Function class base level Meta level 10

Compare Signatures (Input-Output) Evolutionary Programming (R^n -> R) -> R^n Input is a function mapping real-valued vectors of length n to a real-value. Output is a (near optimal) real-valued vector (i.e. the solution to the problem instance) Evolutionary Programming Designer [(R^n -> R)] -> ((R^n -> R) -> R^n) Input is a list of functions mapping real-valued vectors of length n to a real-value (i.e. sample problem instances from the problem class). Output is a (near optimal) (mutation operator for) Evolutionary Programming (i.e. the solution method to the problem class) 11 We are raising the level of generality at which we operate. Give a man a fish and he will eat for a day, teach a man to fish and…

Genetic Programming to Generate Probability Distributions 1.GP Function Set {+, -, *, %} 2.GP Terminal Set {N(0, random)} N(0,1) is a normal distribution. For example a Cauchy distribution is generated by N(0,1)%N(0,1). Hence the search space of probability distributions contains the two existing probability distributions used in EP but also novel probability distributions. CAUCHY GAUSSIAN NOVEL PROBABILITY DISTRIBUTIONS SPACE OF PROBABILITY DISTRIBUTIONS

Ten Function Classes

Parameter Settings Generation and population sizes are low, but we have effectively seeded (or can be easily found) the population with good probability distributions.

Evolved Probability Distributions 1

Evolved Probability Distributions 2

Means and Standard Deviations These results are good for two reasons. 1.starting with a manually designed distributions. 2.evolving distributions for each function class.

T-tests

Evolved Probability Distributions

Differences with Standard Genetic Programming and Function Optimization 1.The final solution is part man-made (the Evolutionary Programming framework) and part machine-made (the probability distributions). 2.We (effectively) seed the initial population with already known good solutions (Gaussian and Cauchy). Don’t evolve from scratch. 3.We train to generalize across specific problem classes, therefore we do not test on single instances but many instances from a problem class.

Further Work 1.Compare with other algorithms (EP with Levy). 2.Only “single humped” probability distributions can be expressed in this framework Consider running GP for longer and stopping automatically (rather than pre-determined) 3.Do not have a single sigma for each automatically designed probability distribution 4.The current framework was sufficient to beat the two algorithms we compared against (Gaussian and Cauchy)

Summary & Conclusions We are not proposing a new probability distribution. We are proposing a method to generate new probability distributions. We are not comparing algorithms on benchmark instances (functions). We are comparing algorithms on distributions. We are using an off-the-shelf method (Genetic Programming) to generate tailor-made solution methods to problem classes.