TUTORIAL DETAILS
T1 - Soft Computing in Data Mining and Knowledge Discovery
T4 - Soft Computing in Bioinformatics
ABSTRACT
Soft computing is a consortium of methodologies that work synergistically
and provides, in one form or another, flexible information processing capabilities
for handling real life ambiguous situations. Fuzzy Sets (FS), Artificial
Neural Networks (ANN), Evolutionary Algorithms (EAs) (including genetic
algorithms (GAs), genetic programming (GP), evolutionary strategies (ES)),
Support Vector Machines (SVM), Wavelets, Rough Sets (RS), Simulated Annealing
(SA), Swarm Optimization (SO), Memetic Algorithms (MA), Ant Colony Optimization
(ACO), Tabu Search (TS), Chaos Theory and Case Based Reasoning (CBR) are
the major components of Soft Computing.
The growth in the amount of data collected and generated has exploded in
recent times with the widespread automation of various day-to-day activities,
advances in high-level scientific and engineering research and development
of efficient data collection tools. This has given rise to the need for
automatically analyzing the data in order to extract knowledge from it,
thereby making the data potentially more useful. Knowledge discovery and
data mining (KDD) is the process of identifying valid, novel, potentially
useful and ultimately understandable patterns from massive data repositories.
It is a multi-disciplinary topic, drawing from several fields including
expert systems, machine learning, intelligent databases, knowledge acquisition,
case based reasoning, pattern recognition and statistics.
Over the past few decades, major advances in the field of molecular biology,
coupled with advances in genomic technologies, have led to an explosive
growth in the biological information generated by the scientific community.
This deluge of genomic information has, in turn, led to an absolute requirement
for computerized databases to store, organize and index the data, and for
specialized tools to view and analyze the data. Bioinformatics can be viewed
as the use of computational methods to make biological discoveries. It is
an interdisciplinary field involving biology, computer science, mathematics
and statistics to analyze biological sequence data, genome content and arrangement,
and to predict the function and structure of macromolecules. The ultimate
goal of the field is to enable the discovery of new biological insights
as well as to create a global perspective from which unifying principles
in biology can be derived.
Two very important problems in the analysis of biological data are superfamily
classification of proteins, and design of ligands for molecular docking.
- Protein superfamilies comprise groups of proteins having similarity
in functions and structures. The importance of super family classification
lies in proper identification of proteins, database maintenance, biological
data mining and identification and proper functional assignment of uncharacterized
proteins with a final goal towards drug discovery and finding homologies.
- The functional part of a protein molecule where various other ligands
may be attached by non-covalent or covalent interactions is known as
active site. Such bindings of ligands in the active site give rise to
the onset of various biochemical pathways inside the body. Identifying/designing
a suitable ligand using groups from a library of chemical groups which
can bind to the active site of a protein to prevent its proliferation
is an important and recent research issue in Bioinformatics.
In the first part of this tutorial we will describe the basic principles
of some soft computing techniques, as well as how these can be utilized
for data mining and knowledge discovery. In this regard, particular emphasis
will be placed on evolutionary algorithms, simulated annealing, fuzzy
theory and neural network. Real-life applications will be demonstrated.
Recently, use of soft computing tools for solving bioinformatics problems
have been gaining the attention of researchers because of their ability
to handle imprecision, uncertainty and near optimality in large and complex
search spaces. The second part of the tutorial will describe the basic
principles of Bioinformatics, with particular emphasis on the problems
of protein classification and drug design. In this regard, application
of the principles of soft computing for solving these two problems will
be discussed in detail.
PRESENTERS
Dr. Ujjwal Maulik did
his BS, MS and Ph.D in Computer Science during 1989, 1991 and 1997 respectively.
He is currently a professor in the Department of Computer Science and
Technology, Jadavpur University, India. Dr. Maulik has worked as a scientist
or visiting professors in Center for Adaptive Systems Application and
Los Alamos National Laboratories, Los Alamos, New Mexico, USA, in 1997,
University of New South Wales, Sydney, Australia in 1999, University of
Texas at Arlington, USA in 2001, University of Maryland at Baltimore in
2004 and AiS laboratory in Fraunhofer Institute, Germany in 2005. Dr.
Maulik is a Fellow of the Institution of Electronics and Telecommunication
Engineers (IETE), India, and a senior member of Institute of Electrical
and Electronics Engineers (IEEE), USA. He has edited several books and
journals and is a co-author of around ninety technical articles in international
journals, book chapters and conference/workshop proceedings. He has served
on the program committees of several International Conferences, and has
delivered many invited talks and tutorials. His research interests include,
Soft Computing, Pattern Recognition, Distributed Data Mining, Bioinformatics
and Parallel and Distributed Systems. He has recently edited a book titled
"Advanced methods for knowledge discovery from complex data", published
by Springer, UK in 2005. He is also editing a book titled "Analysis of
biological data: A soft computing approach" to be published by World Scientific
in 2006.
Dr. Sanghamitra Bandyopadhyay
did her BS, MS and Ph. D. in Computer Science in 1988, and 1991, 1993
and 1997 respectively. Currently she is an associate professor at Indian
Statistical Institute, India. She has worked in Los Alamos National Laboratory,
Los Alamos, USA in 1997 as a graduate research student, in University
of New South Wales, Sydney, Australia, during 1999 as a post doctoral
fellow, in Department of Computer Science and Engineering, University
of Texas at Arlington, USA, as a faculty and researcher in 2001 and in
University of Maryland at Baltimore, USA as a visiting associate professor
during 2004. Dr. Bandyopadhyay is the first recipient of Dr. Shanker Dayal
Sharma Gold Medal and also the Institute Silver Medal for being adjudged
the best all round post graduate performer in IIT, Kharagpur in 1994.
She has also received the Young Scientist Awards of the Indian National
Science Academy (INSA) and the Indian Science Congress Association (ISCA)
in 2000. In 2002 Dr. Bandyopadhyay received the Young Scientist Awards
of the Indian National Academy of Engineers (INAE). Dr. Bandyopadhyay
was an invited speaker at the 8th International Conference on Human and
Computers 2005, held in Aizu, Japan during 30th August to 2nd September
2005. During September 2005 to November 2005 Dr. Bandyopadhyay has worked
as a scientist in the AiS laboratory in Fraunhofer Institute, Germany.
She is a senior member of Institute of Electrical and Electronics Engineers
(IEEE). Dr. Bandyopadhyay has co-authored nearly hundred technical articles
in international journals, book c hapters and conference/workshop proceedings.
She has delivered many invited talks and tutorials. She has recently edited
a book titled "Advanced methods for knowledge discovery from complex data",
published by Springer, UK in 2005, and is preparing an edited book titled
"Analysis of biological data: A soft computing approach" to be published
by World Scientific and an authored book titled "Genetic algorithms for
pattern recognition and learning: Bioinformatics and web intelligence"
to be published by Springer Verlag. She has also edited journals special
issues in the area of Soft Computing, Data Mining and Bioinformatics.
Her research interests include Pattern Recognition, Data Mining, Soft
and Evolutionary Computation, Bioinformatics and Image Processing.
T2 - Optimal Trajectory Planning
of Manipulators
ABSTRACT
Trajectory planning is one of the fundamental issues in the design and
development of manipulators. A time-trajectory may be generated in joint
space or Cartesian space. In joint space trajectories, trajectories are
specified for each independent joint. The actual Cartesian position of
the end-effector is only known at the initial and goal position. On the
other hand, Cartesian trajectories are easy to specify and tip motion
of the manipulator is completely specified. Joint motion is obtained via
the velocity Jacobian. Since trajectories are not generated in joint space,
care must be taken that the trajectories do not pass, or close to, singularities.
Trajectories are chosen to fairly smooth to allow reasonable time for
the manipulator to accelerate and to decelerate. The trajectories in both
joint and Cartesian space schemes can be chosen in a number of ways.
The trajectory is normally determined to satisfy a certain criterion optimally.
Optimal performance means different things to different people such as
minimum time, minimum kinetic energy, and obstacle avoidance. Optimization
is normally performed in the presence of constraints. In addition to the
dynamic system equations acting as constraints, there may be bounds on
the inputs as well as constraints on some of the states. The constraints
are of two types: The system constraints imposed by the manipulator itself
and task constraints given by the task. The problem is how to calculate
feasible trajectories from a given path with simultaneous utilization
of the maximal capabilities of the manipulator.
It is the objective of this tutorial to provide Review, discussion and
analysis of optimization techniques.
TOPICS
The structure of this tutorial will be organized as follows:
1. Basics of Trajectory Selection
1.1 Path versus Trajectory
1.2 Polynomials in time, cubic polynomial, and splines in time
1.3 Linear interpolation with smoothing and linear function with parabolic blends
1.4 Optimal control methods like the shooting method
2. Optimization Techniques
2.1 Constrained and unconstrained optimisation
2.2 Non-stochastic Optimization
2.3 Stochastic Optimization
3. Minimum Time Trajectory Planning
3.1 Cubic-spline versus B-spline
3.2 Kinematic Approach
3.3 Dynamic Approach
3.4 On-line Trajectory generation
4. Minimum kinetic Energy Trajectory Planning
4.1 Mathematical Modelling
4.2 Optimization Criterion
4.3 Case Study
5. Obstacle Avoidance Trajectory Planning
5.1 Briefing on Genetic Algorithms (GAs) technique
5.2 Hybrid Optimization Techniques
5.3 Case study
PRESENTER
Dr. Atef A. Ata received his B. Sc. Degree with Honor in Mechanical
Engineering from Alexandria University in Egypt in 1985. After his graduation he joined the same
university as a Lecturer where he obtained his M. Sc. Degree in Engineering Mathematics
(Hydrodynamics) in 1990. In 1996 he obtained his Ph. D in Engineering Mathematics (Robotics) as
a Joint-Venture between University of Miami, Florida, USA and Alexandria University in Egypt. He
joined Alexandria University again as Assistant Professor till 2001. Then he joined the
Mechatronics Engineering Department, International Islamic University Malaysia as an Assistant
Professor, Associate Professor (2004) and as a Head of Department (2005-now). Dr. Atef is a member
of IEEE Robotics and Automation Society as well as Egyptian Engineering Syndicate. He is also one of the
Editorial Consultant Board for 2006 for the International Journal of Advanced Robotic Systems, Austria.
His research interest includes Dynamic and Control of Flexible Manipulators, Trajectory Planning, Genetic
Algorithms and Modeling and Simulation of Robotic Systems.
T3 - Classification and Decision Making
based on intelligent feature extraction
ABSTRACT
In all learning techniques and decision methods we need to classify various objects, behaviors and
patterns. Classification is a fundamental activity and is at the heart of all decision-making and
AI based processes. In this tutorial we will discuss advanced classification algorithms in detail.
We will talk about various pattern and behavior based methods used for text and image classification
and can be used for various other object classifications. The tutorial will also cover usage of various
methods like SVM, intelligent feature extraction and feature based clustering for these applications.
In this tutorial we will cover various difficulties we face while classifying objects and learning for
the same. We will also study various methods for classification that can handle nonlinear behavior very
effectively. The dimensions of the problem and its impact on performance and accuracy with reference to
various methods in context with curse of dimensionality will be discussed in detail.
We will also elaborate how to extract features from a data set intelligently and how we can use
pattern matching for clustering. How behavioral patterns can be used for classification and what
are these behavior patterns, will also be discussed in detail. We will also explain various learning
philosophies and methods for classification. It will cover supervised and unsupervised learning
methodologies. Supervised and unsupervised learning, Semi-supervised learning and incremental
learning and its applications and impact on classification results will be discussed in detail.
We will also discuss mind-map based decision-making to enable classification the way human being
perceives it. Further how decision objects and action links are generated and time dependant
modeling helps in decision-making will be covered. The tutorial will also present industrial
applications of classification and case studies for classifying images and textual data. It will
also throw some light on how these techniques can be extended for other applications in decision
engineering.
TOPICS
1. Introduction
1.1 What is classification?
1.2 Various classification methods
1.3 Classification and Decision Support systems
1.4 Classification of Knowledge maps and mind maps
2. Intelligent feature extraction for classification
2.1 Behavioral patterns
2.2 Behavioral patterns of images and objects
2.3 Clustering based classification
2.4 Tracking changes in behavior and learning
3. Various methods for classification
3.1 Support vector machines
3.2 SVM for image classification
3.3 SVM for text classification
3.4 Nonlinear behavior pattern and SVM
4. Case studies
4.1 Industrial applications
4.2 Face authentication
4.3 Image based change detection
5. Applications
5.1 Industrial applications
5.2 Decisions based on classifications
5.3 Learning based on classification
6. Concluding remarks
PRESENTER
Dr. Parag Kulkarni is Ph.D. form IIT, Kharagpur. An alumnus of IIT
and IIM Dr. Parag is working in IT industry for more than 15 years. He is on research panel and
Ph.D. guide for University of Pune, BITS and Symbiosis deemed University and guiding 8 Ph.D.
students. He has more than 35 publications and two patents pending in US PTO. He has conducted
more than 10 tutorials at various international conferences and was a keynote speaker for three
international conferences. Recipient of Oriental scholarship he has worked at senior management
positions in Siemens, Ideas. He has developed a new feature based authentication mechanism while
working with Siemens. He has also worked as a referee for International Journal for Parallel and
Distributed Computing, IASTED conferences. He is member of IASTED technical committee of Parallel
and Distributed Computing, WSEAS working committee. Presently he is Chief Scientist and Research
Head at Capsilon India, Pune. He is also Honorary Professor at two prime institutes in Pune and on
board of studies for a couple of Institutes. He is invited to conduct special sessions at IITs', IIMs',
UCC, NICM-Pune, Symbiosis and Pune University. His areas of interest include image processing, security
systems, decision systems, Mind maps, Knowledge maps, expert systems, classification techniques, load
balancing and distributed computing.
T5 - Video Compression Techniques
ABSTRACT
The 'New or Improved' MPEG4 Functionalities
The vision behind the MPEG4 standard is best explained through the eight 'new or improved
functionalities', described in the MPEG4 Proposal Package Description. These eight functionalities
come from an assessment of what will be useful in near future applications, but is not (or only
partly) supported by current coding standards. There are also several other important, so-called
'standard', functionalities, that MPEG4 needs to support as well, just like the already available
standards. Examples are synchronisation of audio and video, low delay modes, and interworking.
Unlike the new or improved functionalities, the standard functionalities may be provided by existing
or emerging standards. The 'new or improved' MPEG4 functionalities are listed below:
- Content-Based Scalability
- Content-Based Manipulation and Bitstream Editing
- Content-Based Multimedia Data Access Tools
- Hybrid Natural and Synthetic Data Coding
- Coding of Multiple Concurrent Data Streams
- Improved Coding Efficiency
- Robustness in Error-Prone Environments
- Improved Temporal Random Access
For each of the functionalities, some examples of their usefulness are suggested.
H.264/MPEG4 part 10
Originally proposed by Toshiba with Main Profile, Level 4.:
I + P picture types, In-loop deblocking,1/4 sample motion compensation, VLC-based entropy coding,
Tree-structured motion segmentation down to 4x4 block size, 4:2:0 Bi-predictive slices, CABAC,
Weighted prediction, Adaptive block-size transforms Interlace pictures, frame/field adaptive at
picture and macroblock level Max bitrate: 29.4Mbps Prediction (GOVU) structure.
Also called JVT o ITU H.264 o ITU-T Rec. H.264 o MPEG-4 Part 10 o ISO 14496-10 AVC Advanced Video
Coding o AVC ...whatever way it may be called, this is revolutionary. H.264 is being widely
recognized as the future platform of video compression for applications such as new HDTV services,
portable game console, mobile broadcast video services, video on solid-state camcorders, instant
video messaging on cell phone. H.264 is the most advanced video coding standard available today.
It uses many new coding techniques not available in MPEG2, MPEG4 and H.263.
PRESENTER
Dr. N. Malmurugan has
around Nineteen years of experience in DSP domain. Have a good Experience
in Audio, Image and Video Compression projects. Worked extensively in Modeling,
Development, and Testing of new image and video compression algorithms using
Wavelet related transforms. Have a good expertise in JPEG, MPEG and H.26x
standards. Before joining Industry, he was in academia and served as Professor
at Dept. of Electronics and Communication Engineering, PSG College of technology,
Coimbatore, India. Currently he heads the DSP division of Satyam Computer
Services Limited, Bangalore, India. He is strong in developing tools and
system applications using C, C++ and MATLAB. He has good knowledge of Multimedia
Applications. His research interests include Wavelet based Signal and Image
processing, Biomedical electronics and Multimedia. He has published more
than 50 papers in National and International level conferences and journals.
He is Fellow of Institute of Electronics and Telecommunication Engineers
( India ), Fellow of Institute of Engineers ( India ) and Fellow of Society
for Simulation and Modeling.
T6 - Structural Learning of Neural Networks
ABSTRACT
The objective of this tutorial is to explain our structural learning methods for neural networks.
(1) Regarding hierarchical neural networks, the number of units for each of input and output is
decided according to problems. But the number of units in a hidden layer cannot be defined easily
and previously. If the small number of units is employed for the hidden layer, the computation time
will be shortened but the optimal solution is not assured. Nevertheless, if a large number of units
are employed for the hidden layer, the huge computation time should be required that the neural
network terminates but frequently it does not assure to obtain the optimal solution. Therefore, we
must fix the most appropriate structure of a neural network for each problem.
Generally, neural network spends much computation time and cost in solving a problem. The reason is
because a neural network requires exponential time in computation according to the number of units
in a hidden layer.
The structural learning is to optimally build a neural network through executing the neural network.
The results enable us to reduce the computational time and cost as well as to understand the simple
structure more easily.
In the tutorial the structural learning method is employed in forecasting the price movement of a
stock. The optimization of the network by the structured learning is evaluated based on its real use.
(2) Regarding a mutual connected neural network, the structure is usually decided by a problem. But
sometimes, it is not given apriori. The decision itself is to solve a problem. We also employ the
structural learning method to solve an integer mixed quadratic programming problem using a mutual
connected neural network.
It is important that the limited amount of investing funds should be efficiently allocated to many
stocks so as to reduce its risk. This problem is formulated as a zero-one mixed integer programming
problem. However, it is not easy to solve the zero-one mixed integer programming problem because of
its combinatorial nature. Therefore, an efficient approximate solution is required to solve a
large--scale zero-one mixed integer programming problem.
In this paper, we propose a meta-controlled Boltzmann machine to obtain an approximate solution of
a large--scale zero-one mixed quadratic programming problem. This model employs a Hopfield network
as the meta-controlling layer to select investing stocks and a Boltzmann machine as the lower layer
to decide investing ratio of each stock. This model deletes the units of the lower layer
corresponding to units which are not selected in the meta-controlling layer in its execution.
Then the lower layer is restructured by using the selected units. Executing the meta-controlled
Boltzmann machine according to the above mentioned algorithm, the meta-controlled Boltzmann
machine converges more efficiently than a conventional Boltzmann machine. In this paper, we
evaluate efficiency of a meta-controlled Boltzmann machine employing various sizes of data.
(3) One application is exemplified for each structural learning method. The effects are discussed.
You can employ the structural learning method in your neural network. You can develop a new horizon in a research.
PRESENTER
Prof. Dr. Junzo Watada was born at Osaka in 1945.
He received his B.S. and M.S. degrees in electrical engineering from Osaka City University, Japan, and Dr. of Eng. degree through the research on fuzzy multivariate analysis from Osaka Prefecture University, Japan. He is a Professor of Knowledge Engineering, Soft Computing and Management Engineering at the Graduate School of Information, Production & Systems, the Waseda University since 2003, after a professor of Human Informatics and Knowledge Engineering, the School of Industrial Engineering at the Osaka Institute of Technology, Japan. Before moving to Academia, he was with Fujitsu Co. Ltd. as a senior systems engineer for 7 years. His research interests are broad including fuzzy system methodologies, automata theory, text and web mining, grid computer systems, decision support systems and experts systems, DNA computer, security system, data analysis, financial engineering, macro-ergonomics, etc.. Recently he works actively on image processing, intelligent systems, genetic algorithms, neural networks , DNA computing, and so on. He was the President of Bio-Medical Fuzzy Systems Association (2001-2003). He was
the Vice President of Japan Society for Fuzzy Theory and Systems for two years (1993-1995) and was a
board committee of Japan Society for Fuzzy Theory and Systems, and also serve as an advisory board member
for several international and domestic societies and also an editorial board member for international and
domestic journals. Dr. Watada received
- The Contribution Award at ISIS2002 in Korea on August 25, 2001,
- Henri Coanda Gold Medal in Romania on July 17, 2002,
- Excellent Presentation Award of SCIS2002 at Tsukuba on October 21-25, 2002,
- Board of Certification in Professional Ergonomics as Certified Professional Ergonomist by Japan Ergonomics Society on August 3, 2003,
- Contribution Award, Biomedical Fuzzy Systems Association, November, 2004,
- Fellow, Society of Japan Intelligent Informatics and Fuzzy Systems, January 10, 2005, and
- The Contribution Award, International Anniversary Symposium "Grigore C. Moisil" SASM2005, May 1-3, 2005
- The Contribution Award, Japan Society of Fuzzy Theory and Intellectual Informatics, September 10, 2005,
- The Contribution Award to developing fuzzy systems, on behalf of Professor L.A. Zadeh, BISC, Special Event 40 years of Fuzzy Systems, November 3, 2005.
|
