Social Network Analysis

Login |
 
 

Lecture: Factorization Models for Machine Learning

Abstract

Factorization models play an important role in many applications, among them e.g. recommender systems, text or network analysis. Recently, factorization models have gained lot of interest and research in the field of machine learning because of their high quality predictions, e.g. winning the $1,000,000 Netflix challenge. In contrast to classical factorization approaches such as singular value decomposition (SVD) or principal component analysis (PCA), in machine learning factorization models are treated differently, esp. in learning and regularization. This lecture will cover the basic matrix factorization model, higher-order factorization models as well as hierarchical approaches. The second focus is on learning methods which covers iterative approaches (e.g. gradient descent) and Bayesian inference (Markov Chain Monte Carlo).

All models and methods will be introduced from scratch -- prior knowledge about factorization models e.g. SVD, PCA is not necessary.

Overview

SWS4
Credits6
LanguageEnglish
LocationC424
TimeTue, 13:30-15:00
Thu, 15:15-16:45
InstructorSteffen Rendle

Topics

  • Introduction: Factorization Models and the Netflix Prize
  • Matrix Factorization
  • Learning and Regularization
  • Bayesian Inference
  • Higher-order factorization models
  • Hierarchical models
  • Collective models
  • Applications
All lecture material can be found in ILIAS.
(Mathematisch-Naturwissenschaftliche Sektion - Informatik und Informationswissenschaft - Lehrveranstaltungen SS 13 - Factorization Models for Machine Learning)

Credit Requirements

  • oral exam on Mon, 15th July 2013.
  • 2nd exam on Mon, 14th October 2013. Please write to steffen.rendle (AT) uni-konstanz.de to get your examination time.
  • each exam takes 30 minutes