COSC Seminar Series - November 7.
The following seminar(s) are planned for this Thursday November 7, 2-3pm, J328. They will be given by 2 visiting graduate students from the University of Pretoria (South Africa), who work with Dr Andries P. Engelbrecht.
(1) Heterogeneous Particle Swarms in Dynamic Environments
Heterogeneous particle swarms allow particles to use different
velocity- and position update equations from one another. This work
investigates the performance and scalability of heterogeneous particle
swarms in dynamic environments. The results are compared to that of
charged particle swarms (CPSO) and quantum particle swarms (QPSO). It
is shown that some heterogeneous swarms are able to manage the
diversity of the swarm dynamically, allowing it to overcome the
problem of diversity loss and to successfully track a moving optimum
over time. Additionally, findings show that dHPSO scales well to high
severity and high frequency DF1 environments. For MPB environments,
similar scalability results are observed, with dHPSO obtaining better
average results over all test cases.
(2) Self-Adaptive Heterogeneous Particle Swarms
Particle swarm optimisation is a stochastic optimisation technique in which a
swarm of particles moves around a d-dimensional hyper-cube with the aim of
optimising a function. The movement of the particles is controlled by the
personal experience of each particle and the social interaction within
the swarm. Generally, the movement of each particle is performed in the same manner as
the other particles in the swarm. This limits the search behaviour exhibited
by the individual particles and, collectively, that of the swarm.
Heterogeneous particle swarm optimisers allow the particles to use different behaviours
from each other. Dynamic and adaptive heterogeneous particle swarm optimisers
allow the particles to change their behaviours during the search process.
The thesis investigates the behaviour changing aspects of these heterogeneous
particle swarm optimisers. Three self-adaptive heterogeneous particle swarm
optimisation variants are proposed along with a number of behaviour changing
schedules. A comparison is performed on the different behaviour selection
processes and the scalability to large scale optimisation functions is investigated.