Drukuj Eksport do pliku (MS Word)
Nazwa przedmiotu:
Pattern Recognition
Koordynator przedmiotu:
mgr inż. Rajmund Kożuszek
Status przedmiotu:
Fakultatywny dowolnego wyboru
Poziom kształcenia:
Studia II stopnia
Program:
Informatyka
Grupa przedmiotów:
Przedmioty techniczne - zaawansowane
Kod przedmiotu:
EPART
Semestr nominalny:
4 / rok ak. 2015/2016
Liczba punktów ECTS:
6
Liczba godzin pracy studenta związanych z osiągnięciem efektów uczenia się:
30 hours of lectures 24 hours preparation for tests 15 hours of laboratory exercises 20 hours of preparation for the laboratory exercises 15 hours of project meetings 40 hours of implementation of project assignments
Liczba punktów ECTS na zajęciach wymagających bezpośredniego udziału nauczycieli akademickich:
30 hours of lecture 15 hours of laboratory exercises 15 hours of project meetings which gives approx. 2.5 ECTS
Język prowadzenia zajęć:
angielski
Liczba punktów ECTS, którą student uzyskuje w ramach zajęć o charakterze praktycznym:
15 hours of laboratory exercises 20 hours of preparation for laboratory exercises 15 hours of project meetings 40 hours of implementation of project assignments which gives approx. 4 ECTS
Formy zajęć i ich wymiar w semestrze:
  • Wykład30h
  • Ćwiczenia0h
  • Laboratorium15h
  • Projekt15h
  • Lekcje komputerowe0h
Wymagania wstępne:
basic knowledge of linear algebra and probability theory
Limit liczby studentów:
24
Cel przedmiotu:
The aim of the course is to introduce students to the issue of pattern classification. During lectures will be discussed: general structure of image recognition systems, selected methods and techniques of classification and problems related to data collection, image segmentation and dimensionality reduction. The examples of applications of these methods in recognition systems (especially implemented by students during laboratory exercises and project assignments) will allow the students to analyze the practical aspects of image recognition.
Treści kształcenia:
Introduction: the components of the image recognition system; classifier design cycle; quality of classifiers and classification systems assessment methods. Optimal Bayesian classification: the role of a priori information; the form of a probability density function; Bayesian optimal classifier; consideration of the risks and losses when constructing a classifier; decision boundaries of classifiers; compliance data distribution with the adopted theoretical distribution. Nearest neighbor methods: template matching; minimum-distance classifiers; metrics; k-NN classifiers; nearest neighbor search methods; nearest neighbor search acceleration; editing and reduction of the training set. Linear classifiers: linear classification functions; homogeneous space; determining the decision boundary: support vector machine; sequential minimal optimization algorithm. Dimensionality reduction: Principal Component Analysis; Fisher's linear classification; multivariate discriminant analysis (MDA). Clustering: definition of clustering problem; assessment of the similarity of clustering; k-means algorithm; bottom-up clustering; graph algorithms; density based clustering (DBSCAN). Neural networks: the basic model of the neuron; training algorithms for a single neuron; interpretation of a single neuron operation; neural networks; error backpropagation algorithm; network with feedback loops; associative memory; Hopfield memory; Kohonen self-organizing network; ART networks. Markov models: discrete Markov processes; hidden Markov process; forward and backward procedures; Viterbi algorithm; Baum-Welsha procedure; using Markov models in classification. Text searching: exact and approximate text matching; Boyer-Moore algorithm; edit distance; text analysis using non-deterministic and deterministic automata; suffix tree and suffix array; Ukkonen algorithm for constructing the suffix tree; approximate search with suffixe tree; neighborhood generation for search with errors; hash functions for fast searching. Decision trees: construction of decision trees: basic CART algorithm; assessment of heterogenity of tree nodes; stop criteria when generating the tree; the effect of the horizon; tree pruning algorithms. Improving the quality of classification: basic problems of constructing metaclassifiers; majority voting; the independence of the classifiers; weighted voting; determination of weights; Bayesian methods for combining of classification results; Behavior-Knowledge Space; bagging and boosting (AdaBoost algorithm); the use of contextual information in the classification; context in OCR systems; use of trigrams and dictionaries.
Metody oceny:
two tests (on the 7th and penultimate lecture) 3 laboratory exercises (unrated introduction, 2 exercises rated on a scale 0-6) 2 mini-projects (4 project meetings) rated on a scale 0-12
Egzamin:
nie
Literatura:
Duda R.O., Hart P.E., Stork D.G., Pattern Classification, Wiley-Interscience, 2000 Stąpor K., Automatyczna klasyfikacja obiektów, Exit, Warszawa 2005 Jain A. K., Fundamentals of Digital Image Processing, Prentice-Hall International Editions, Engelwood Hills, 1989 Tadeusiewicz R., Korohoda P., Komputerowa analiza i przetwarzanie obrazów, Wydawnictwo Fundacji Postępu Telekomunikacji, Kraków 1997 Press W. H., Numerical Recipes in C, Cambridge University Press, Cambridge 1992 (lub późniejsze wydania)
Witryna www przedmiotu:
https://studia.elka.pw.edu.pl/priv/13Z/EPART.A
Uwagi:
.

Efekty uczenia się

Profil ogólnoakademicki - wiedza

Efekt EPART_W01
Student knows basic pattern classification methods
Weryfikacja: test, laboratory excercises, project assignments
Powiązane efekty kierunkowe: K_W12, K_W06, K_W08, K_W09
Powiązane efekty obszarowe: T2A_W08, T2A_W04, T2A_W07, T2A_W03
Efekt EPART_W02
Student knows preliminary data analysis and clustering methods
Weryfikacja: test, laboratory excercises, project assignments
Powiązane efekty kierunkowe: K_W06, K_W08, K_W09
Powiązane efekty obszarowe: T2A_W04, T2A_W07, T2A_W03
Efekt EPART_W03
Student knows basic classifiers ensamble construction methods
Weryfikacja: test, laboratory excercises, project assignments
Powiązane efekty kierunkowe: K_W12, K_W08
Powiązane efekty obszarowe: T2A_W08, T2A_W07

Profil ogólnoakademicki - umiejętności

Efekt EPART_U01
Student can analyze a training set, design a simple classifier and evaluate its quality
Weryfikacja: laboratory excercises, project assignments
Powiązane efekty kierunkowe: K_U01, K_U06
Powiązane efekty obszarowe: T2A_U01, T2A_U08, T2A_U09
Efekt EPART_U02
Student can, on the basis of a training set assessment, choose a classification method and compute its parameters
Weryfikacja: test, laboratory excercises, project assignments
Powiązane efekty kierunkowe: K_U01, K_U06, K_U07
Powiązane efekty obszarowe: T2A_U01, T2A_U08, T2A_U09, T2A_U10
Efekt EPART_U03
Student is able to critically evaluate the solution of the classification problem and propose its improvement
Weryfikacja: test, laboratory excercises
Powiązane efekty kierunkowe: K_U07, K_U10, K_U11
Powiązane efekty obszarowe: T2A_U10, T2A_U15, T2A_U16