X1. Formal methods in AI
Lecturers: Tinko Tinchev, Dimiter Vakarelov
Study Hours: 4+0

The course is not an introduction in classical and nonclasical logics, neither it is a course on logic programming. The main goal of this course is to create a basis for a more solid understanding of various notions in AI such as knowledge representation, common sense reasoning, reasoning under incomplete information and so on.

The topics selected in this course and their organization conform to the idea to demonstrate the advantages of integrating the language of logic with AI.

The exercises are selected with the aim of more practical work on the language of first order predicate calculus especially on the deductive methods presented in the course.

X2. Natural Language Processing
Lecturers: Galia Angelova
Study Hours: 3+1

The main goal of the lectures is to give a general view of the field of computational linguistics, within 30 academic hours. Areas like e.g. Machine translation and Text corpora are briefly mentioned. Some other fields will be considered in more detail: the students will be introduced to the basic language structures and methods for their processing. The main topics include: morphology, in classificational and two-level models, with emphasis on Bulgarian morphology; syntax and unification formalisms; natural language generation.

The main goal of the seminars is to give the students a practical view on the problems of computer morphology, syntax and NL generation. A computer morphology of Bulgarian will be considered, as well as a HPSG-based parser of simple Bulgarian sentences. The system DB-MAT for generation of NL explanations (see http://www.informatik.uni-hamburg.de/NATS/projects/db-mat.html) will be used as a framework for practical exercises in NL Generation and lexicons of NLP systems. The basic programming language to be used during the seminars is Prolog.

L1. AI Programing in Prolog
Lecturers: Svetla Boycheva
Study Hours: 1+0+5

The course is a basic introduction to artificial intelligence (AI) covering introductory material in searching, game playing, problem solving, knowledge representation, deduction, planning, machine learning, expert systems, and natural-language processing. The course will focus on some specific programming techniques and methods (in Prolog), used in AI.

Contents

L2. Programming on Internet
Lecturers: Sergei Varbanov
Study Hours: 1+0+5

The course is a basic introduction to HTML and Java Script aplications building.

Z1.Machine Learning
Lecturers: Svetla Boycheva
Study Hours: 2+0+2

The course gives a broad overview of the ML approaches and further focuses on some of them mostly within the Inductive paradigm. Some basic algorithms for Unsupervised learning are also discussed. The inductive learning approaches are emphasised since they are natural extensions and applications of classical AI techniques. Thus the course can easily fit in an AI curriculum.

The description of the material uses some fundamental notions of First Order Logic and Logic Programming (Predicate calculus, Resolution etc.), which are common for any AI course. The exercises are based mainly on Prolog. Therefore a general AI course and a Prolog programming course are necessary to be taken before attending the ML course.

The exercises are based on programming ML techniques in Prolog and implementing some simple ML algorithms. Several ready-to-use ML systems are provided within Prolog framework, such as ID3, COBWEB and FOIL. They are used for experiments and for further elaboration in advanced projects.

Z2. Graphical Image Recognition - methods, algorithms, tools
Lecturers: Dimo Dimov
Study Hours: 2+0+2

Graphics Recognition (GR), Optical Character Recognition (OCR), and in particular Handwriting Word Recognition (HWR), are well known as Document Analysis (DA) domains of Informatics. Despite of its recent advance, they are considered still open for new ideas and research. Two cases of technologies can be generally met here: (i) "off-line" case where the input (text and/or graphics) is considered a final 2/D image, and (ii) "on-line" case when the text information is being obtained simultaneously with its drawing. Each minimal image of isolated text herein, words in the HWR and isolated letters in the "classical" OCR, can be treated as separate graphics that leads to many analogies between GR and HWR/OCR. These analogies can be met at the initial stages of the image pre-processing and of the primary recognition by parts as well, while the specifics of application (cartography, banking, criminology, etc.) dominate at the final stages of a DA system. Any case, relevant strings of letters over an alphabet (or structures of strings) are expected on the output. Combination of probability and structural methods of recognition as well as reference of lexicons and usage of feed-backs is considered very prospective for a recognition plausibility improvement recently.

The course objectives are to represent general methods, algorithms and tools for a DA system development.

Main topics:

1. Image processing and pattern recognition. Basic notions and problems. Recognition of 2d images. Document analysis
2. Tools for image input/output. Computer graphics and image analysis duality. Graphic file formats
3. Digitalization. Fourier transform. Shannon’s theorem. Raster and vector representations.
4. Image enhancement. Linear and nonlinear filters. Segmentation.
5. Feature selection from grey-scale images. Finding the contours. Skeletonization, thinning, vectorization.
6. Registration of specific graphical elements in images. Hough transform for line segments. Generalized Hough transform.
7. Feature selection for classification. Linear feature spaces. Kahrunen-Loeve transform.
8. Image transforms. Pyramids. Wave transform.
9. Elastic matching of graphical images. Template – image correlation.
10. Stochastic methods for recognition and classification. Bayes theory. Supervised and unsupervised learning, parametric and nonparametric approach. Self-teaching.
11. Building classifiers by neural networks. Nonlinear multilayer perceptron. Associative Hopfield memory.
12. Structural and syntactical methods for graphics and text recognition. Using formal languages and grammars for 2d image description.
13. Graphics recognition by approximate string matching. Recognition improvement through a template dictionary.
14. Image data bases. Data bases for document analysis. Methods for efficient access.
15. Information technology for graphics and text recognition. The recognition process can be considered as a program system with feedbacks.

References

1. D. Ballard, C. Brown, "Computer vision", Prentice hall inc., NY , 1982.
2. G. A. Baxes, "Digital image processing – principles and applications, J. Wiley Sons, NY 1994.
3. Fundamentals in handwriting recognition, In S. Impedovo (ed.), NATO ASI Series F, vol.124, Springer-Verlag, Berlin 1994.
4. T. Masters, "Signal and image processing with neural networks (a C++ source book), J. Wiley Sons, NY 1994.
5. Progress in handwriting recognition, A.C.Downtown and S. Impedovo (eds.) , World scientific, 1997.
6. Optical character recognition, Proceedings of the IEEE, T. Pavlidis and S.Mori (eds.), 80(7), July 1992.

 

Z3.Cognitive Science
Lecturers: Boycho Kokinov
Study Hours: 4+0

Z4.Case-Based Reasoning (second year)
Lecturers: Genady Agre
Study Hours: 2+0+2

The course is aimed at computer science students at postgraduate level (M.Sc.). It is intended to provide the students with the background knowledge in the field of case-based reasoning (CBR) - a new paradigm for combining problem-solving and learning. The different aspects of case-based reasoning process will be illustrated by examples of CBR systems which use past experience to solve problems in different problem domains. Some basic Instance-Based (IBL) and Exemplar-Based Learning algorithms which are frequently used for implementing CBR systems will be also described.

The students will have a possibility to run Prolog and LISP implementations of simplified versions of some CBR systems described in the course as well as to experiment with the presented basic IBL and EBL algorithms.

Course Syllabus

Z5. Principles of Knowledge-Based Systems (second year)
Lecturers: Maria Nisheva
Study Hours: 2+0+2

The course covers the main characteristics and architectural principles of knowledge-based systems
(KBS) and the up-to-date methods for building KBS. Several popular knowledge representation
languages (OPS5, KL-ONE, KRL) and tools for building KBS are introduced. The basic methods
for uncertain knowledge representation are discussed. Some topics concerning knowledge
representation and reasoning by analogy are involved. The problems of truth maintenance systems
development are briefly discussed. Students are assigned a small project that consists in building a
rule-based system using the CLIPS system as a tool.

Basic Program of the Course:

   1.Basic notions and types of knowledge in KBS, Principles of KBS Architecture.
   2.Analysis of popular formalisms and languages for knowledge representation and processing. Dealing with uncertainties. Keeping
     the robustness
   3.Models for knowledge representation in the systems for classification, diagnosis and problem solving.
   4.Problem solving by analogy.
   5.Getting knowledge in KBS. Generation of explanations.
   6.Instrumental tools for creating KBS.

References:

   1.Brachman, R., H. Levesque (Eds.). Readings in Knowledge Representation. Morgan Kaufmann, 1985.
   2.Forgy, C. RETE - A Fast Algorithm for the Many Pattern/Many Object Pattern Match Problem. Artificial Intelligence, No. 19, 1982,
     pp. 17-37.
   3.Jackson, P. Introduction to Expert Systems (2nd ed.). Addison-Wesley, 1990.
   4.Michalski, R., J. Carbonell, T. Mitchell (Eds.). Machine Learning. Tioga, CA, 1983.
   5.Michalski, R., J. Carbonell, T. Mitchell (Eds.). Machine Learning, Vol. 2. Morgan Kaufmann, 1986.
   6.Rich, E., K. Knight. Artificial Intelligence (2nd ed.). McGraw-Hill, 1991.
   7.Sowa, J. (Ed.). Principles of Semantic Networks: Explorations in the Representation of Knowledge. Morgan Kaufmann, 1991.
   8.Stefik, M. Introduction to Knowledge Systems. Morgan Kaufmann, 1995.
 

Z6. Neurual Networks (second year)
Lecturers: Sergei Varbanov
Study Hours: 2+0+2

Z7. Motion Planning in Cluttered Environment (second year)
Lecturers: Antony Popov
Study Hours: 2+0+2

     This lecture course considers the problems of off-line and on-line motion planning in cluttered  environment with static and dynamic obstacles. The problem of planning the co-operative actions of  multiple moving agents is considered as well. Specific techniques and algorithms from computational   geometry are introduced for modelling the working environment and for establishing simple criteria  for collision detection.  As applications,     the following problems are discussed:
     - Some basic notions: 3D geometric modelling by primitives, configuration space, hidden elements removal,  Gilbert’ s algorithm for distance calculation.
    - The abstract robot as a generalization of any moving agent.
    - Collision -free motion planning of robot manipulators.
    - Planning the co-operative actions of two moving objects  in the presence of static   obstacles.
    - Distribution of the access to a common data base in a multi-user system.
    - Planning the trajectory of a sensor-guided mobile robot.

     Practical workshops using simulation software such as TRUCKWORLD are provided. Some analogs between 3D computer  graphics and workspace modelling, for instabce Binary Space Partitions (BSP), painter’s algorithm. Some preliminary knowledge from the kinematics of a rigid body will be discussed, like Denavit- Hartenberg formalism, which is useful also for 3D graphics and computer animation.

Literature:

1. K. Fujimura, Motion planning in dynamic environment, Springer-Verlag,  Tokyo, 1991
2. P. H. Winston, Artificial intelligence (3ed.), Addison-Wessley, 1992
3. J.C. Latombe, Robot motion planning, Kluwer, Norwell-MA, 1991
 
 
 

Z8. Declarative Languages for Knowledge Representation (second year)
Lecturers: Kiril Simov
Study Hours: 2+0+2