Xin Yao: "What can evolutionary computation do for you?"

What Can Evolutionary
Computation Do For You?

Xin Yao (http://www.cs.bham.ac.uk/~xin)
CERCIA, School of Computer Science, University of Birmingham, UK

http://www.cercia.ac.uk

Er …

 Nothing (if you are not in this field)?
 A lot (if you are in this field)?
 Sometimes it can be very useful (for
most people).
 But what “times”?
 Maybe we should look at a broader
picture …

What Existing Computers
Still Can’t Do Well

 Brittle
 Doesn’t learn
 Hopeless in dealing with noisy and
inaccurate information
 Never grow up
 Unable to adapt
 Turning us into slaves
 but also give us jobs …

What Is Evolutionary Computation

 It is the study of computational systems
which use ideas and get inspirations
from natural evolution.
 Evolutionary computation (EC)
techniques can be used in optimisation,
learning and creative design.
 EC techniques do not require rich
domain knowledge to use. However,
domain knowledge can be incorporated
into EC techniques.

Outline of This Talk

 Adaptive Optimisation
 Data Mining and Machine Learning

Case Study I:
Modelling: Both Structures
and Parameters

Discovering `New’ Physical Laws in
Astrophysics --- Modelling Radial Brightness
Distributions in Elliptical Galaxies

 Empirical laws are widely used in astrophysics.

 However, as the observational data increase,
some of these laws do not seem to describe
the data very well.

 Can we discover new empirical laws that
describe the data better?

Monochromatic (Negative) Images of Galaxies

bulge disk

A typical elliptical galaxy A typical spiral galaxy

Current Approach

 Select a functional form in advance
Drawbacks: ad hoc, difficult to determine and may only suit a
smaller number of profiles

 Apply fitting algorithms to find suitable
parameters for the function. Usually adopt
the non-linear reduced c2 minimization
1 [ I mo d el (i )  I o b s (i )] 2
c2 =

 i 2
where Iobs(i) is the individual observed profile value, Imodel(i) is the value calculated from the
fitting function,  is the number of degrees of freedom, and  is the standard deviation of the
data.
Drawbacks: difficult to set initial values and easily trapped in local minima

Our Evolutionary Approach

(1) Find functional forms using GP (Genetic Programming) :
 A data-driven process without assuming a functional form in
advance
 A bottom up process which suits modelling a large number of
galaxy profiles without any prior knowledge of them
(2) Fit parameters in the form found using FEP (Fast
Evolutionary Programming):
 Not sensitive to initial setting values
 More likely to find global minima

J. Li, X. Yao, C. Frayn, H. G. Khosroshahi, S. Raychaudhury, ``An Evolutionary Approach to
Modeling Radial Brightness Distributions in Elliptical Galaxies,'' Proc. of the 8th
International Conference on Parallel Problem Solving from Nature (PPSN VIII), Lecture
Notes in Computer Science, Vol. 3242, pp.591-601, Springer, September 2004.

Case Study II:
Material Modelling

Determining Unified Creep Damage Constitutive
Equations in Aluminium Alloy Modelling

 Creep behaviours of different materials are
often described by physically based unified
creep damage constitutive equations.
 Such equations are extremely complex.
 They often contain undecided constants
(parameters).
 Traditional approaches are unable to find good
near optima for these parameters.
 Evolutionary algorithms (EAs) have been shown
to be very effective.

Evolutionary Parameter
Calibration and Optimisation

 Classical evolutionary programming
(CEP), fast EP (FEP) and improved
fast EP (IFEP) can be used to find
parameters in a complex and non-
differentiable space.

B. Li, J. Lin and X. Yao, “A novel evolutionary algorithm
for determining unified creep damage constitutive
equations,” International Journal of Mechanical
Sciences, 44 (2002): 987–1002.

Model with Evolved Parameters by IFEP

Real World Impact

 Used by Corus / Tata Corus.

Case Study III:
Constraint Handling

Constraint Handling

 Real-world problem has many
constraints, e.g., linear, nonlinear,
equality, inequality, …

Evolutionary Constraint Handling

It works better than other methods because
 More effective;
 Good at dealing with non-differentiable
and nonlinear problems;
 Avoid unnecessary and unrealistic
assumptions.

Stochastic Ranking

 It is a simple yet effective constraint
handling method.
 It exploits the characteristics of
evolutionary algorithms.

 T. P. Runarsson and X. Yao, “Stochastic Ranking for Constrained
Evolutionary Optimization,” IEEE Transactions on Evolutionary
Computation, 4(3):284-294, September 2000.
 T. Runarsson and X. Yao, “Search Bias in Constrained
Evolutionary Optimization,” IEEE Transactions on Systems, Man,
and Cybernetics, Part C, 35(2):233-243, May 2005.

Impact Outside Evolutionary
Computation

 It’s used so often that people made it a C library
so that everyone can use it easily:
 Xinglai Ji 1 and Ying Xu 2, “libSRES: a C library for stochastic
ranking evolution strategy for parameter estimation”
Bioinformatics, 22(1):124-126, 2006.
 1Computational Biology Institute, Oak Ridge National Laboratory Oak
Ridge, TN 37831, USA
2Department of Biochemistry and Molecular Biology and Institute of

Bioinformatics, University of Georgia Athens, GA 30602-7229, USA

 Also in civil, mechanical, electrical and electronic
engineering.

Real World Is Dynamic

 Real world is complex and changes
all the time.
 E.g., customers make new orders or cancel existing
orders all the time. How can I optimise my objectives?

Case Study IV:
Dynamic Optimisation

Route Optimisation for Gritting Trucks

 Optimisation Problem:
 Multiple trucks with
different capacities
 Multiple depots
 Complex routing
constraints
 Dynamic
 Evolutionary algorithms
outperformed all other
existing algorithms

Further Information

 Application:
 H. Handa, L. Chapman and Xin Yao, ``Robust route
optimisation for gritting/salting trucks: A CERCIA
experience,'' IEEE Computational Intelligence
Magazine, 1(1):6-9, February 2006.
 Theory:
 P. Rohlfshagen, P. K. Lehre and X. Yao, “Dynamic
evolutionary optimisation: An analysis of frequency
and magnitude of change,” In Proceedings of the
2009 Genetic and Evolutionary Computation
Conference, pages 1713-1720, 2009. (best paper
award)

Case Study V:
Multi-objective Optimisation

Ubiquity of Multi-objective
Problem Solving: in Hardware

 A practical example:
 X. Yao and T. Schnier, “HARDWARE DESIGN
USING EVOLUTION ALGORITHMS,” European
Patent EP1388123. Held by Ericsson (formerly
Marconi).
 Underpinning research:
 T. Schnier, X. Yao and P. Liu, ``Digital filter design
using multiple pareto fronts,'' Soft Computing,
8(5):332-343, April 2004.

Ubiquity of Multi-objective Problem
Solving: in Software

 K. Praditwong, M. Harman and X. Yao, “Software
Module Clustering as a Multi-Objective Search
Problem,” IEEE Transactions on Software
Engineering, 37(2):264-282, March/April 2011.
 Z. Wang, K. Tang and X. Yao, “Multi-objective
Approaches to Optimal Testing Resource
Allocation in Modular Software Systems,” IEEE
Transactions on Reliability, 59(3):563-575,
September 2010.

Outline

 Adaptive Optimisation

 Data Mining and Machine Learning

Evolutionary Computation Helps
Population-based Thinking

 Such a new way of thinking can lead
to interesting connections to other
research fields, e.g., machine learning.
 One example is ensemble learning, i.e.,
“a population of heads are better than
one”:
 X. Yao and Y. Liu, “Making use of population
information in evolutionary artificial neural
networks,” IEEE Trans. on Systems, Man, and
Cybernetics, Part B: Cybernetics, 28(3):417-425,
June 1998.

Case Study VI:
Ensemble Learning

Negative Correlation Learning

 Negative Correlation Learning is a very
simple ensemble learning algorithm
inspired by evolutionary computation.
 Y. Liu and X. Yao, ``Ensemble learning via
negative correlation,'' Neural Networks,
12(10):1399-1404, December 1999.
 It has many applications.
 Xin Yao, Gavin Brown, Bernhard Sendhoff
and Heiko Wersing, “Exploiting ensemble
diversity for automatic feature extraction,”
European Patent EP1378855. (Held by
Honda).

Multi-objective Ensemble
Learning

 Multiple objectives can be easily
accommodated in ensemble learning:
 H. Chen and X. Yao, ``Multiobjective Neural Network
Ensembles based on Regularized Negative
Correlation Learning,'' IEEE Transactions on
Knowledge and Data Engineering, 22(12):1738-1751,
December 2010.
 A. Chandra and X. Yao, “Ensemble learning using
multi-objective evolutionary algorithms,” Journal of
Mathematical Modelling and Algorithms, 5(4):417-445,
December 2006.
 Ensembles + Multi-objective + Cloud = Piero

Many More Applications

 The six case studies presented represent only
a tiny set of all applications of EC.
 Applications are interesting, but where are the
challenges for the future?

Challenge: Theoretical Foundations

 Theories of evolutionary computation have
lagged behind applications, although recent
progresses in computational time complexity
analysis have been substantial.
 We need to understand when an EA is expected
to perform well on what types of problems.
 What makes a problem hard/easy for an EA?

Challenge: Scalability

 At present, large scale optimisation in EC deals
with one or two thousand variables at most.
We need efficient EAs that can optimise
problems with tens or even hundreds of
thousands of variables.

Challenge: Dealing with
Dynamics and Uncertainty

 We do not have a static and predictable world.
Problems are changing even when they are
being solved. Constraints are changing with
time too.
 How can we optimise more effectively and
efficiently in such a dynamic and uncertain
environment?

Challenge: Learn to Optimise

 Optimisation has been treated mathematically.
It has been very rare for any optimisation
algorithm to consider its previous experience
in solving other, often similar, problems.
 Why not? We, human beings, do learn to solve
a problem better by applying previous
experiences.

Enjoy Some of the Challenges?

 Why not join the Birmingham team?
 Available now:
 Fully funded PhD in evolutionary
computation
 Two job vacancies (postdoctoral research
fellows): 3 and 4 years, respectively. Both in
evolutionary computation.
 Also available:
 MRes in Natural Computation

Conclusions

 Evolutionary computation has both
interesting applications and challenging
theories.
 It can do far more than one might initially
have thought.
 There are golden opportunities in the
research and applications of evolutionary
computation.

Xin Yao: "What can evolutionary computation do for you?"

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Xin Yao: "What can evolutionary computation do for you?" (20)

More from ieee_cis_cyprus (7)

Xin Yao: "What can evolutionary computation do for you?"