Information-theoretic methods for data exploration

Lecturers

Tijl De Bie, Ghent University, department of Electronics and information systems

Jeffrey Lijffijt, Ghent University, department of Electronics and information systems. .

Schedule and place

This 15-hour course will take place at KU Leuven

Dates to be determined

Travel instructions are available here

Description

Abstract:

A key problem in the design of methods for exploring data is the quantification of how interesting a 'pattern'  (i.e. a formally described nugget of information) found in data is to the data analyst. This lecture series will describe a framework, called FORSIED  (Formalizing Subjective Interestingness in Exploratory Data mining), for achieving this. The lectures will describe the framework along with a range of instantiations therefor for a broad range of data and problem types.

 

The FORSIED framework, a high-level overview

   * The FORSIED framework

   * Community detection

   * Perspectives for privacy and fairness

 

Exploration of discrete data

   * Itemset mining

   * Relational pattern mining

   * Cycles, trees, and forests in graphs

   * Exploring attributed graph data

 

Exploration of real-valued data

   * Projections

   * Subgroup discovery

   * ct-SNE

   * Time series

 

Graphs and graph embeddings

   * Advanced MaxEnt models

   * CNE

   * ExplaiNE

   * EvalNE

   * DeBayes

   * ALPINE

   * FONDUE

Course material

To be determined

Evaluation

Homeworks/projects