ALAR
The Artificial Life and Adaptive
Robotics Laboratory
1.
|
Abstract |
Dealing
with optimization problems with more than one objective has been an important
research area in evolutionary computation. The class of multi-objective
problems (MOPs) is an important one because
multi-objectivity exists in almost all aspects of human life; whereby there
usually exist several compromises in each problem. Multi-objective
evolutionary algorithms (MOEAs) have been applied
widely in many real-world problems. This is because (1) they work with a
population during the course of action, which hence offer more flexible
control to find a set of efficient solutions, and (2) real-world problems are
usually black-box where an explicit mathematical representation is unknown. However,
MOEAs usually require a large amount of
computational effort. This is a substantial challenge in bringing MOEAs to practice. This thesis primarily aims to address
this challenge through an investigation into issues of scalability and the
balance between exploration and exploitation. These have been outstanding
research challenges, not only for MOEAs, but also
for evolutionary algorithms in general. A
distributed framework of local models using explicit niching
is introduced as an overarching umbrella to solve multi-objective
optimization problems. This framework is used to address the two-part
question about first, the role of communication messages and second, the role
of explicit niching in distributed evolutionary
multi-objective optimization. The concept behind the framework of local
models is for the search to be conducted locally in different areas of the
decision search space, which allows the local models to be distributed on
different processing nodes. During the optimization process, local models
interact (exchange messages) with each other using rules inspired from Particle
Swarm Optimization (PSO). Hence, the hypothesis of this work is that running
simultaneously several search engines in different local areas is better for
exploiting local information, while exchanging messages among those diverse
engines can provide a better exploration strategy. For this framework, as the
models work locally, they gain access to some global knowledge of each other. In
order to validate the proposed framework, a series of experiments on a wide
range of test problems was conducted. These experiments were motivated by the
following studies which in their totality contribute to the verification of
our hypothesis: (1) studying the performance of the framework under different
aspects such as initialization, convergence, diversity, scalability, and
sensitivity to the framework's parameters, (2) investigating interleaving
guidance in both the decision and objective spaces, (3) applying local models
using estimation of distributions, (4) evaluating local models in noisy
environments and (5) the role of communication messages and explicit niching in distributed computing. The
experimental results showed that: (1) the use of local models increases the
chance of MOEAs to improve their performance in
finding the Pareto optimal front, (2) interaction strategies using PSO rules
are suitable for controlling local models, and that they also can be coupled
with specialization in order to refine the obtained non-dominated set, (3)
estimation of distribution improves when coupled with local models, (4) local
models work well in noisy environments, and (5) the communication cost in
distributed systems with local models can be reduced significantly by using
summary information (such as the direction information naturally determined
by local models) as the communication messages, in comparison with
conventional approaches using descriptive information of individuals. In
summary, the proposed framework is a successful step towards efficient
distributed MOEAs.
|
|
Files |
01front.pdf 139.6 Kb |
2. Ang
Yang (PhD), 2003-2007. A networked multi-agent combat model: emergence
explained.
|
Abstract |
Simulation
has been used to model combat for a long time. Recently, it has been accepted
that combat is a complex adaptive system (CAS). Multi-agent systems (MAS) are
also considered as a powerful modelling and development environment to
simulate combat. Agent-based distillations (ABD) - proposed by the US Marine
Corp - are a type of MAS used mainly by the military for exploring large
scenario spaces. ABDs that facilitated the analysis and understanding of
combat include: ISAAC, EINSTein, MANA, CROCADILE
and BactoWars. With new concepts such as networked
forces, previous ABDs can implicitly simulate a networked force. However, the
architectures of these systems limit the potential advantages gained from the
use of networks. In this thesis, a novel network centric multi-agent
architecture (NCMAA) is pro-posed, based purely on network theory and CAS. In
NCMAA, each relationship and interaction is modelled as a network, with the
entities or agents as the nodes. NCMAA offers the following advantages: 1. An
explicit model of interactions/relationships: it facilitates the analysis of
the role of interactions/relationships in simulations; 2. A mechanism to
capture the interaction or influence between networks; 3. A formal real-time
reasoning framework at the network level in ABDs: it interprets the emergent
behaviours online. For a long time, it has been believed that it is hard in
CAS to reason about emerging phenomena. In this thesis, I show that despite
being almost impossible to reason about the behaviour of the system by
looking at the components alone because of high nonlinearity, it is possible
to reason about emerging phenomena by looking at the network level. This is
undertaken through analysing network dynamics, where I provide an
English-like reasoning log to explain the simulation. Two implementations of
a new land-combat system called the Warfare Intelligent System for Dynamic
Optimization of Missions (WISDOM) are presented. WISDOM-I is built based on
the same principles as those in existing ABDs while WISDOM-II is built based
on NCMAA. The unique features of WISDOM-II include: 1. A real-time network
analysis toolbox: it captures patterns while interaction is evolving during
the simulation; 2. Flexible C3 (command, control and communication) models; I
3. Integration of tactics with strategies: the tactical decisions are guided
by the strategic planning; 4. A model of recovery: it allows users to study
the role of recovery capability and resources; 5. Real-time visualization of
all possible information: it allows users to intervene during the simulation
to steer it differently in human-in-the-loop simulations. A comparison
between the fitness landscapes of WISDOM-I and II reveals similarities and
differences, which emphasise the importance and role of the networked
architecture and the addition of strategic planning. Lastly but not least,
WISDOM-II is used in an experiment with two setups, with and without
strategic planning in different urban terrains. When the strategic planning
was removed, conclusions were similar to traditional ABDs but were very
different when the system ran with strategic planning. As such, I show that
results obtained from traditional ABDs - where rational group planning is not
considered - can be misleading. Finally, the thesis tests and demonstrates
the role of communication in urban ter-rains. As
future warfighting concepts tend to focus on
asymmetric warfare in urban environments, it was vital to test the role of
networked forces in these environments. I demonstrate that there is a phase
transition in a number of situations where highly dense urban terrains may
lead to similar outcomes as open terrains, while medium to light dense urban
terrains have different dynamics |
|
Files |
01front.pdf 273.1 Kb |
3.
|
Abstract |
Artificial
neural networks have been widely used for machine learning and optimization.
A neuro ensemble is a collection of neural networks
that works cooperatively on a problem. In the literature, it has been shown
that by combining several neural networks, the generalization of the overall
system could be enhanced over the separate generalization ability of the
individuals. Evolutionary computation can be used to search for a suitable
architecture and weights for neural networks. When evolutionary computation
is used to evolve a neuro ensemble, it is usually
known as evolutionary neuro ensemble. In most
real-world problems, we either know little about these problems or the
problems are too complex to have a clear vision on how to decompose them by
hand. Thus, it is usually desirable to have a method to automatically
decompose a complex problem into a set of overlapping or non-overlapping
sub-problems and assign one or more specialists (i.e. experts, learning
machines) to each of these sub-problems. An important feature of neuro ensemble is automatic problem decomposition. Some neuro ensemble methods are able to generate networks,
where each individual network is specialized on a unique sub-task such as
mapping a subspace of the feature space. In real world problems, this is
usually an important feature for a number of reasons including: (1) it provides
an understanding of the decomposition nature of a problem; (2) if a problem
changes, one can replace the network associated with the sub-space where the
change occurs without affecting the overall ensemble; (3) if one network
fails, the rest of the ensemble can still function in their sub-spaces; (4)
if one learn the structure of one problem, it can potentially be transferred
to other similar problems. In this thesis, I focus on classification problems
and present a systematic study of a novel evolutionary neuro
ensemble approach which I call cooperative coevolutionary
mixture of experts (CCME). Cooperative coevolution
(CC) is a branch of evolutionary computation where individuals in different
populations cooperate to solve a problem and their fitness function is
calculated based on their reciprocal interaction. The mixture of expert model
(ME) is a neuro ensemble approach which can
generate networks that are specialized on different sub-spaces in the feature
space. By combining CC and ME, I have a powerful framework whereby it is able
to automatically form the experts and train each of them. I show that the
CCME method produces competitive results in terms of generalization ability
without increasing the computational cost when compared to traditional training
approaches. I also propose two different mechanisms for visualizing the
resultant decomposition in high-dimensional feature spaces. The first
mechanism is a simple one where data are grouped based on the specialization
of each expert and a color-map of the data records
is visualized. The second mechanism relies on principal component analysis to
project the feature space onto lower dimensions, whereby decision boundaries
generated by each expert are visualized through convex approximations. I also
investigate the regularization effect of learning by forgetting on the
proposed CCME. I show that learning by forgetting helps CCME to generate neuro ensembles of low structural complexity while
maintaining their generalization abilities. Overall, the thesis presents an
evolutionary neuro ensemble method whereby (1) the
generated ensemble generalizes well; (2) it is able to automatically
decompose the classification problem; and (3) it generates networks with
small architectures. |
|
Files |
01front.pdf 229.0 Kb |
4.
Yin Shan (PhD), 2002-2005. Program
distribution estimation with grammar models.
|
Abstract |
This
thesis studies grammar-based approaches in the application of Estimation of
Distribution Algorithms (EDA) to the tree representation widely used in
Genetic Programming (GP). Although EDA is becoming one of the most active
fields in Evolutionary computation (EC), the solution representation in most
EDA is a Genetic Algorithms (GA) style linear representation. The more
complex tree representations, resembling GP, have received only limited
exploration. This is unfortunate, because tree representations provide a
natural and expressive way of representing solutions for many problems. This
thesis aims to help fill this gap, exploring grammar-based approaches to
extending EDA to GP-style tree representations. This thesis firstly provides
a comprehensive survey of current research on EDA with emphasis on EDA with
GP-style tree representation. The thesis attempts to clarify the relationship
between EDA with conventional linear representations and those with a
GP-style tree representation, and to reveal the unique difficulties which
face this research. Secondly, the thesis identifies desirable properties of
probabilistic models for EDA with GP-style tree representation, and derives
the PRODIGY framework as a consequence. Thirdly, following the PRODIGY
framework, three methods are proposed. The first method is Program Evolution
with Explicit Learning (PEEL). Its incremental general-to-specific grammar
learning method balances the effectiveness and efficiency of the grammar
learning. The second method is Grammar Model-based Program Evolution (GMPE).
GMPE realises the PRODIGY framework by introducing elegant inference methods
from the formal grammar field. GMPE provides good performance on some
problems, but also provides a means to better understand some aspects of conventional
GP, especially the building block hypothesis. The third method is Swift GMPE
(sGMPE), which is an extension of GMPE, aiming at
reducing the computational cost. Fourthly, a more accurate Minimum Message
Length metric for grammar learning in PRODIGY is derived in this thesis. This
metric leads to improved performance in the GMPE system, but may also be
useful in grammar learning in general. It is also relevant to the learning of
other probabilistic graphical models. |
|
Files |
01front.pdf 132.3 Kb |
5.
Xuan
Hoai Nguyen (PhD), 2001-2005. Flexible representation
for genetic programming: lessons from natural language processing.
|
Abstract |
This
thesis principally addresses some problems in genetic programming (GP) and
grammar-guided genetic programming (GGGP) arising from the lack of operators
able to make small and bounded changes on both genotype and phenotype space.
It proposes a new and flexible representation for genetic programming, using
a state-of-the-art formalism from natural language processing, Tree Adjoining
Grammars (TAGs). It demonstrates that the new
TAG-based representation possesses two important properties: non-fixed arity and locality. The former facilitates the design of
new operators, including some which are bio-inspired, and others able to make
small and bounded changes. The latter ensures that bounded changes in
genotype space are reflected in bounded changes in phenotype space. With
these two properties, the thesis shows how some well-known difficulties in
standard GP and GGGP tree-based representations can be solved in the new
representation. These difficulties have been previously attributed to the treebased nature of the representations; since TAG
representation is also tree-based, it has enabled a more precise delineation
of the causes of the difficulties. Building on the new representation, a new
grammar guided GP system known as TAG3P has been developed, and shown to be
competitive with other GP and GGGP systems. A new schema theorem, explaining
the behaviour of TAG3P on syntactically constrained domains, is derived.
Finally, the thesis proposes a new method for understanding performance
differences between GP representations requiring different ways to bound the search space, eliminating the effects of the
bounds through multi-objective approaches. |
|
Files |
01front.pdf 100.0 Kb |
6.
Jason Teo (PhD), 2001-2003. Pareto
multi-objective evolution of legged embodied organisms.
|
Abstract |
The
automatic synthesis of embodied creatures through artificial evolution has become
a key area of research in robotics, artificial life and the cognitive
sciences. However, the research has mainly focused on genetic encodings and
fitness functions. Considerably less has been said about the role of
controllers and how they affect the evolution of morphologies and behaviors in artificial creatures. Furthermore, the
evolutionary algorithms used to evolve the controllers and morphologies are
pre-dominantly based on a single objective or a weighted combination of
multiple objectives, and a large majority of the behaviors
evolved are for wheeled or abstract artifacts. In
this thesis, we present a systematic study of evolving artificial neural
network (ANN) controllers for the legged locomotion of embodied organisms. A
virtual but physically accurate world is used to simulate the evolution of
locomotion behavior in a quadruped creature. An
algorithm using a self-adaptive Pareto multi-objective evolutionary
optimization approach is developed. The experiments are designed to address
five research aims investigating: (1) the search space characteristics
associated with four classes of ANNs with different
connectivity types, (2) the effect of selection pressure from a self-adaptive
Pareto approach on the nature of the locomotion behavior
and capacity (VC-dimension) of the ANN controller generated, (3) the effciency of the proposed approach against more
conventional methods of evolutionary optimization in terms of computational
cost and quality of solutions, (4) a multi-objective approach towards the
comparison of evolved creature complexities, (5) the impact of relaxing
certain morphological constraints on evolving locomotion controllers. The
results showed that: (1) the search space is highly heterogeneous with both
rugged and smooth landscape regions, (2) pure reactive controllers not
requiring any hidden layer transformations were able to produce sufficiently
good legged locomotion, (3) the proposed approach yielded competitive
locomotion controllers while requiring significantly less computational cost,
(4) multi-objectivity provided a practical and mathematically-founded
methodology for comparing the complexities of evolved creatures, (5)
co-evolution of morphology and mind produced significantly different creature
designs that were able to generate similarly good locomotion behaviors. These findings attest that a Pareto
multi-objective paradigm can spawn highly beneficial robotics and virtual
reality applications. |
|
Files |
01front.pdf 150.6 Kb |
Last update: Sept 20, 2007
02:01:05 PM ALAR: Artificial Life
and Adaptive Robotics Lab