Spring 2016 / DM843
Unsupervised Learning
General Information
One trend can be observed over almost all fields of informatics: we have to cope with an ever-increasing amount of available data of all kinds. This amount of data renders it impossible to inspect the dataset "by hand", or even deduce knowledge from the given data, without sophisticated computer aided help. In this course we will discuss one of the most common mechanism of unsupervised machine learning for investigating datasets: Clustering. Clustering separates a given dataset into groups of similar objects, the clusters, and thus allows for a better understanding of the data and their structure. We discuss a number of clustering methods and their application to various different fields such as biology, economics or sociology.
Procedure of the oral exam
The exam will last about 25 minutes. At the beginning, one topic from the list below will be drawn randomly. For each topic the examinee should be prepared to make a short presentation of about 8 minutes. It is allowed to bring one page of hand-written notes (DIN A4 or US-Letter, one-sided) for each of the topics. The examinee will have 2 minutes to briefly study the notes for the drawn topic before the presentation. The notes may be consulted during the exam if needed but it will negatively influence the evaluation of the examinee's performance. During the presentation, only the blackboard can be used (you cannot use overhead transparencies, for instance).
After the short presentation, additional question about the presentation's topic but also about other topics in the curriculum will be asked.
Below is the list of possible topics and some suggested content. The listed content are only suggestions and is not necessarily complete nor must everything be covered in the short presentation. It is the responsibility of the examinee to gather and select among all relevant information for each topic from the course material. On the course website you can find suggested readings for each of these topics.
Topics for the Oral Exam:
- Graphical Analysis of Clusters
- Histograms, Scatter Plots
- Density Estimation (Parzen Windows, Kernel vs. kNN)
- Principal Component Analysis
- Principal Coordinate Analysis
- Proximity Measures
- Different datatypes
- Common measures for various data types
- Metrics
- Similarities for structural data
- Hierarchical Clustering
- Function principles / linking functions
- Dendrograms
- Comparison to crisp clusterings
- Function principle of BIRCH
- Optimization Based Clustering
- Dissection of the co-variance matrix (W and B)
- Cluster criteria
- K-means
- GAP statistic
- Cluster Analysis
- Steps of a cluster analysis
- Data preprocessing
- Cluster Validation (internal vs. external)
- Expectation Maximization
- Function principle
- Gaussian Mixture Models
- Maximum Likelihood Estimators
- Similarity to k-means
- Advanced Clustering
- Subspace clustering
- Ensemble clustering
- Co-clustering
Student Talks
You will receive one or two scientific papers about the clustering tool assigned to you. You are supposed to create a small presentation which should be 15-20 minutes. It is important that you stay within this time frame! In your small talk, you should cover the following aspects:
- Present the underlying idea and the algorithm of the tool
- How was the tool evaluated (e.g., on what datasets, compared against what other tools, formal proofs of certain properties, etc.)
- What is your opinion about the pros and the cons of the algorithm
The paper is only the starting point and you should use this as the basis for your presentation, but feel free to dig-up any other sources.
The talks will be on the 9th and 10th of May. Please take into account, that this talk is mandatory to be eligible for the exam. In case you have not yet registered for a talk, please write me an email and I will assign you a topic.