Sie sind nicht angemeldet

Unsupervised Machine Learning


Dozent/in MSc, Sandro Cilurzo, MSc, Arthur Habicht;
Veranstaltungsart Vorlesung
Code FS251028
Semester Frühjahrssemester 2025
Durchführender Fachbereich Wirtschaftswissenschaften
Studienstufe Master
Termin/e Mi, 26.02.2025, 16:15 - 20:00 Uhr, E.508
Mi, 12.03.2025, 16:15 - 20:00 Uhr, E.508
Mi, 26.03.2025, 16:15 - 20:00 Uhr, E.508
Mi, 09.04.2025, 16:15 - 20:00 Uhr, E.508
Mi, 07.05.2025, 16:15 - 20:00 Uhr, E.508
Mi, 21.05.2025, 16:15 - 20:00 Uhr, HS 2
Umfang 2 Semesterwochenstunden
Turnus Bi-weekly
Inhalt
Machine learning algorithms can be separated on a high level in two fundamental different types - supervised and unsupervised. Supervised machine learning algorithms are better known to the general public in comparison to unsupervised approaches. Classifying breast cancer on images which have been annotated by doctors can be seen as one real-world example of supervised machine learning. Supervised machine learning algorithms can be extremely powerful but are often limited by the availability of labeled data. Tedious and costly manual labor is necessary to prepare data sets which can be fed into supervised machine learning algorithms to achieve the expected performance. On the other hand, unsupervised machine learning algorithms are meant to find structures and relationships in the raw data itself, without any labels or prior information provided by human supervisors. 
This course will introduce several unsupervised machine learning techniques which can be leveraged in different domains - from finding hidden structures in time series data, representing text information in a numerical way until possibilities of generating new image data. To achieve all of that, we will introduce algorithm by algorithm in a rigorous manner guided by examples. The participants will learn when and how an unsupervised machine learning technique could be applicable. Furthermore, they will be able to implement them by themselves and expand their data analysis tools at their disposal.  
Summarized goals and scope: 
- understand the difference of unsupervised machine learning and supervised machine learning
- clustering (K-means, DBSCAN, agglomerative clustering)  
- dimensionality reduction (robust pca, t-SNE)
- semi-supervised machine learning algorithms
         o introduction to autoencoders and their applications (e.g. automated feature engineering)
         o word2vec algorithm to generate numerical embeddings of textual data
- generative models
         o discriminative vs generative models
         o creating images with variational autoencoders
Lernziele - deep understanding of the benefits and limitations of different learning paradigms in machine learning
- an overview of different unsupervised machine learning techniques to solve different classes of problems (time series data, textual data & images)
- developing an intuition about composition possibilities of using several machine learning algorithms at once
- relationship between the curse of dimensionality and lower dimensional representations
- automated feature engineering and its pros and cons
- generative vs discriminative models
- gather hands-on experience in leveraging unsupervised machine learning algorithms in code
- the participants are expected to be able to create code implementations by themselves
- understand the impact of different parameterizations for each showed algorithm
Voraussetzungen - working experience with Python and its most important tools (pip, virtualenv etc.)
- statistical foundations
- willingness and eagerness to learn
- tinkering mindset
Sprache Englisch
Begrenzung Max. 25 participants
Anmeldung

To attend the course / exercise, registration via e-learning platform OLAT is required. Registration is possible from 3 – 28 February 2025. The students themselves are responsible for checking the creditability of the course to their course of study.

Direct link to OLAT course: https://lms.uzh.ch/url/RepositoryEntry/17670242379

Prüfung ***IMPORTANT*** In order to acquire credits, resp. to take the examination, registration via the Uni Portal within the examination registration period is ESSENTIALLY REQUIRED. Further information on registration: www.unilu.ch/wf/pruefungen
Abschlussform / Credits Written report / 4.5 Credits
Hörer-/innen Nach Vereinbarung
Kontakt sandro.cilurzo@sedimentum.com
arthur.habicht@sedimentum.com
Anzahl Anmeldungen 0 von maximal 25
Literatur Deep Learning Book (Ian Goodfellow)