Clustering is a basic tool used in data analysis, pattern recognition and machine learning for finding groups in data. K-means is still the most popular algorithm in clustering. But is it good enough? How to decide how many clusters? What if the data is non-numerical like categorical, graph, text or more complex objects like GPS trajectories? Outliers, noise and missing values also degrade the clustering performance so how to deal with these problems? Besides these problems, clustering problem is like other optimization problems. It consists of the following three main design problems: (1) define distance function suitable for the data, (2) select cost function to measure goodness of the clusters, (3) design algorithm to optimize for the cost function.
Lecturer: Pasi Fränti
Course assistants: Sami Sieranoja and Gulraiz I Choudhary
Video lectures (~28h): Thursday 14-16 (Teams)
Exercises (7): Tuesdays 14-16 (Teams)
Starting from 12.1.2021 (Intro lecture)
12.1. Discussion of practicalities
13.1. Introduction to clustering
20.1. K-means, Fast k-means, Random swap
3.2. Graph clustering, Mumford-Shah k-means
10.2. Cost functions, text clustering, clustering of web pages
17.2. Clustering evaluation, outlier detection
24.2. Number of clusters, location-based data
3.3. Divisive clustering, Genetic algorithm
8.3. Density peaks, case study
10.3. Agglomerative clustering (on-line + discussion)
All lectures in YouTube
Exercise 1: 18.1.
Exercise 2: 1.2.
Exercise 3: 8.2.
Exercise 4: 15.2.
Exercise 5: 22.2.
Exercise 6: 1.3.
Exercise 7: 8.3.
Submit your exercises in Moodle
Design & Analysis of Algorithms
18.3. 12-16, Room M100 (Joensuu), Room CA101 (Kuopio)
22.4. 12-16, Room M100 (Joensuu), Room CA101 (Kuopio)
Lectures Notes and material from 2014