HK92 + H_x0008_K93 Genetic Algorithms and

advertisement
H02D1A
Genetic Algorithms and Evolutionary Computing
Dirk Roose
1d semester; 4 credit points
Aims






to describe genetic algorithms and (other) evolutionary strategies for
search and optimisation
to analyse their performance (quality of results, computational cost)
to discuss some implementation issues
to illustrate the methods by solving some model problems (e.g.
travelling salesman problem, transportation problem)
to present some case studies (e.g. concept learning, timetabling,
‘artificial life’)
the student will be able to decide whether these methods are suited to
solve a particular search or optimisation problem, and how to choose
the appropriate methods / genetic operators
Genetic Algorithms and Evolutionary Computing
COMPUTATIONAL
INTELLIGENCE
or
SOFT COMPUTING
Neural
Networks
Evolutionary
Programming
Evolutionary
Algorithms
Evolution
Strategies
Fuzzy
Systems
Genetic
Algorithms
Genetic
Programming


Prerequisite
basic (bachelor) courses in informatics (programming, algorithms)
and mathematics (analysis, statistics)
Course Material
– some chapters from Genetic Algorithms and Genetic Programming.
Modern Concepts and Practical Applications. M. Affenzeller, S.
Winkler, S. Wagner, and A. Beham, Chapman and Hall/CRC 2009
(book ; e-book)
– some papers

Teaching activities
– lectures : 12 x 1.5 hour
last lectures: short presentations by students about ongoing project
(incl. interesting or unexpected results, questions)
– exercises & practical sessions : 4 x 2.5h = 10 h

Project
experiments with Matlab code, groups of 2 students, ± 40 hours

Exam
open book exam (theory & exercises) incl. discussion on project report
Genetic Algorithms and Evolutionary Computing

Lectures :
Wednesday 9 am
Celestijnenlaan 200D (Physics building), room 05.11

Exercises:
2 groups; on Mondays & Tuesdays
H03F9A
Parallel computing
Dirk Roose & Albert-Jan Yzelman
1d semester
Aims
Insight in
 parallel computers and available
software environments,
 the design and performance analysis
of parallel algorithms.
KU Leuven HPC Cluster with 2736 ’cores’
The student will be able to
 design efficient parallel versions of algorithms with simple data
dependencies
 both in the ‘shared address space’ programming model and in the
‘message passing’ programming model.
Revolution is Happening Now



Chip density is
continuing to increase
~2x every 2 years
– Clock speed is not
– Number of
processor cores
may double
instead
There is little or no
hidden parallelism
(ILP) to be found
Parallelism must be
exposed to and
managed by software
Source: Intel, Microsoft (Sutter) and
Stanford (Olukotun, Hammond)
6
Parallel computing: Content

Architecture of parallel HPC systems (short)

Performance analysis on parallel systems

Design and analysis of parallel algorithms for model problems
(matrix operations, sorting, fast Fourier transform) using the BSP model

Simple examples in BSPlib, BSPonMPI, MulticoreBSP
(MPI: Message Passing Interface)

Dynamic load-balancing

Guest lecture on Parallel Matching by Rob Bisseling

…

Parallel computing
Prerequisites
Bachelor-level knowledge of algorithms & programming
Course material
Rob H. Bisseling, Parallel Scientific Computation.
A Structured Approach using BSP and MPI.
Oxford University Press, 2004.
+ some papers
Exercises and practical sessions
5 sessions (3 on parallel systems:
multicore processor; HPC-cluster)
no project
Exam
Open book exam
– insight in theory (in particular performance analysis)
– design of an efficient parallel algorithm (high level description)
Exascience Lab
see www.exascience.com
9
Download