Bo#omUp Subspace Clustering
A Symbio9c Approach
Ali Vadhat and Malcolm Heywood
The Subspace Clustering Task
Lance Parsons et al., (2004)Subspace
Clustering for High Dimensional Data: A
review.
SigKDD Explora9ons.
Optimization of function y = x2 GA Example by Hand
From Genetic Algorithms in Search, Optimization and Machine Learning. Goldberg 1989
1/13/03 2  Simple GA Example 1 1/13/03
1000 900 800 700 600 500 400 300 200 100 0 0 5 10 15 20 25 30 35 x over interval
CSCI6506 A little Machine Learning 101
Instructor: Malcolm Heywood
September 7, 2011
1. What is Machine Learning (ML) and how does it relate to Articial
Intelligence (AI) as a whole?
2. How is a problem posed to a ML algorithm to solve?
3. How do the basi
From Genetic Algorithms
to Genetic Programming
Malcolm Heywood
CSCI6506
The Evolutionary
Computation Domain
Learning Classifier
Systems
Genetic Algorithms
Optimization style
representation
Solution is a number
Evolving a rule base
Genetic Programmin
CSCI6506 Genetic Algorithm and Programming
Hollands GA Schema Theorem
Objective provide a formal model for the effectiveness of the GA search process. In the following we will first approach the problem through the framework formalized by Holland [1] and
Active Learning in GP:
The Dynamic Subset Selection
Family
Malcolm Heywood
Context
Computational expense of inner loop
Case of multiclass classification
#Evals = #Class #Trials #Gen Pop_Size  TD 
where
#Class Num. Classes  1
#Trials Num. Populatio
Classification as Clustering:
A Pareto CooperativeCompetitive
GP Approach
Andrew McIntyre, Malcolm Heywood
Evolutionary Computation Journal
(Spring 2011)
MIT Press
Classification Problem Domain
c dimensional class
Consistent space
Initially
Unknown mapp
Symbiosis as a Mechanism for
building Complex Adap9ve Systems:
A Review
Malcolm Heywood and Peter
Lichodzijewski
Dalhousie University, Computer Science
tosynthetic
plankton in
rth and are
cers in the
eyesPrieto
kowski and
tence)
Symbiosis, Complexica2on and
Simplicity under GP
Peter Lichodzijewski
and Malcolm Heywood
Dalhousie University, Computer Science
tosynthetic
plankton in
rth and are
cers in the
eyesPrieto
kowski and
tence) varieties do frequently oc
Nave Crossover Biases in GP
On the Evolution of Parsimonious
Solutions
20 March 2k3
CSCI 6506
1
Case for Parsimonious
Solutions
Avg(program size increase)
O (generation2)
Evaluation Overhead Increases
Solution Transparency Decreases
Does boat more effect

cfw_
z
y
l
g
s
r
x
t
w
s
v
u
t
s
r
j
q
h
k
p
o
m
n
m
l
k
j
i
h
g
f
f
e
d
~
v
u
p

cfw_

~

cfw_
s
z
q
t
y
x
v
w
v
u
t
s
r
q
p
o
o
)
(
'
&
%
$
#
"
!
n
h
h
m
l
w
t
s
r
What are scheduling problems GA scheduling
From Practical Issues and Recent Advances in Job and OpenShop Scheduling; Corne D., Ross P.
2/3/03 GA Scheduling 1
Allocation of tasks to a limited set of resources such that the measure of schedule quality is