Although complex Machine Learning models may provide accurate predictions, they often suffer from being biased or difficult to explain. In this seminar, we will explore current approaches for more transparency, explainability and fairness in different areas of Machine Learning and Data Analytics.

General Information

Lecturer: Prof. Gjergji Kasneci; Organisation: Martin Pawelczyk, Tobias Leemann

The registration and organisation of the seminar will be handled via Ilias. Please sign up early as the number of places is limited.

Our first meeting will take place on Monday, 24th of October 2022 at 2:00pm in room A104 (Sand 14). There we will share further information about the content and organisation of the seminar. Usually, students are required to complete small research projects in teams up to two.

Prerequesites: There are no specific prerequisites. A solid background in Computer Science and Mathematics, statistics in particular, is beneficial.

Topics

Explainability for Online Machine Learning

• Developing Explanation Methods and Properties for Dynamic Data Streams
  • Can our understanding of explainability change over time? Is post-hoc and local explainability meaningful and feasible in a data stream?
• Explainable Online Learning with Decision Rules
  • Are algorithms based on decision rules more suited for explainable online learning than post-hoc attribution methods?

Conceptual Explanations

• Extending Annotated Concepts with Learned Concepts for Complete Explanation
  • How can automatically annotated concepts be extended with meaningful unsupervisedly discovered concepts?
• Explaining Decisions in Terms of Connected Concepts
  • Suppose we have a graph of concepts and their semantic connection (and maybe some form of representation such as a GNN). How can we explain what a GNN has learned and come up with some kind of attributions?

Counterfactual Explanations and Strategic Classification

• How can adversarial attacks impact counterfactual explanations?
• How robust are sequential counterfactual explanations?
• Can we devise strategy-robust counterfactual explanations?
• Are counterfactual explanations highlighting the most important features?

Fairness for automated decision-making systems

• Evaluating fairness for automated decision making systems (especially risk and welfare considerations).
  • We want to study how the convex formulation of welfare-based measures of fairness allows us to integrate them as a constraint into any convex loss minimization pipeline.

Explainability of deep neural networks through feature attribution.

• Evaluation of local explanations

Although complex Machine Learning models may provide accurate predictions, they often suffer from being biased or difficult to explain. In this seminar, we will explore current approaches for more transparency, explainability and fairness in different areas of Machine Learning and Data Analytics.

General Information

Lecturer: Prof. Gjergji Kasneci; Organisation: Vadim Borisov, Martin Pawelczyk, Johannes Haug, Tobias Leemann

The registration and organisation of the seminar will be handled via Ilias. Please sign up early as the number of places is limited.

Our first meeting will take place on Monday, 25th of April 2022 at 10:00am at Sand in room A104. There we will share further information about the content and organisation of the seminar. 

Prerequesites: There are no specific prerequisites. A solid background in Computer Science and Mathematics, statistics in particular, is beneficial.

Topics

Explainability for Online Machine Learning

• Developing Explanation Methods and Properties for Dynamic Data Streams
  • Can our understanding of explainability change over time? Is post-hoc and local explainability meaningful and feasible in a data stream?
• Explainable Online Learning with Decision Rules
  • Are algorithms based on decision rules more suited for explainable online learning than post-hoc attribution methods?

Conceptual Explanations

• Extending Annotated Concepts with Learned Concepts for Complete Explanation
  • How can automatically annotated concepts be extended with meaningful unsupervisedly discovered concepts?
• Explaining Decisions in Terms of Connected Concepts
  • Suppose we have a graph of concepts and their semantic connection (and maybe some form of representation such as a GNN). How can we explain what a GNN has learned and come up with some kind of attributions?

Counterfactual Explanations and Strategic Classification

• How can adversarial attacks impact counterfactual explanations?
• How robust are sequential counterfactual explanations?
• Can we devise strategy-robust counterfactual explanations?
• Are counterfactual explanations highlighting the most important features?

Fairness for automated decision-making systems

• Evaluating fairness for automated decision making systems (especially risk and welfare considerations).
  • We want to study how the convex formulation of welfare-based measures of fairness allows us to integrate them as a constraint into any convex loss minimization pipeline.

Explainability of deep neural networks through feature attribution.

• Evaluation of local explanations

This lecture gives an introduction into probability theory, statistics, information theory, data (pre-)processing and indexing techniques, efficient similarity analysis, as well as linear and nonlinear predictive models. Additionally, the relevant concepts of explainability and fairness in machine learning will be discussed.

The lecture will be held in english. Please register to the lecture via Ilias (first come, first serve).

Information

Lecturer: Gjergji Kasneci; Exercises: Johannes Haug, Vadim Borisov, Martin Pawelczyk, Tobias Leemann.

Although complex Machine Learning models may provide accurate predictions, they often suffer from being difficult to explain. In this seminar, we will explore current approaches for more transparency and explainability in different areas of Machine Learning and Data Analytics.

Information

Lecturer: Prof. Gjergji Kasneci; Organisation: Vadim Borisov, Martin Pawelczyk, Johannes Haug

Please use ILIAS to register and receive updates on the organisation and content of the seminar (the number of participants will be limited to 30 students). The seminar can also be found in Alma. 

Our first meeting will take place on Monday, 26th of April 2021 at 12:00pm (noon) via Zoom (a link will be shared on Ilias). There we will share further information about the content and organisation of the seminar. 

The lecture gives an introduction into probability theory, statistics, information theory, data (pre-)processing and indexing techniques, efficient similarity analysis, as well as linear and nonlinear predictive models. In addition to last year's lecture, we also introduce relevant concepts of fairness and explainability in machine learning.

The lecture will be held in english and online via Zoom. The number of participants is limited to 70 in order to enable a written examination with physical presence under the current circumstances. Please register to the lecture via Ilias (link below; first come, first serve).

Information

Lecturer: Gjergji Kasneci; Exercises: Johannes Haug, Vadim Borisov, Martin Pawelczyk: Ilias

Although well-trained complex Machine Learning models may provide accurate predictions, they often suffer from being difficult to explain. In this seminar, we will explore how interpretablity and transparency for such models can be achieved, while simultaneously producing human-understandable and useful explanations.

Information

Lecturer: Prof. Gjergji Kasneci; Organisation: Martin Pawelczyk, Hamed Jalali, Johannes Haug

More information about the seminar can be found in Alma. Use Ilias to register. 

Preliminary Schedule & Setup

The intital meeting takes place on Monday, 27. April 2020. We plan to conduct the meeting online, using zoom, microsoft teams or another online communication platform. Please use Ilias to register for the course since we will disseminate all required information using this platform. In order to successfully take part in the seminar, we ask students to complete small research projects. In June, before the exam period starts, students are asked to present their findings. The first meeting provides more information about the details of the research projects, supervision and organizational aspects. 

The lecture gives an introduction into probability theory, statistics, information theory, data (pre-)processing and indexing techniques, efficient similarity analysis, classification, clustering and probabilistic inference in graphical models. The lecture and exercise sessions will be held in english.

Information

Lecturer: Gjergji Kasneci; Exercises: Johannes Haug, Vadim Borisov, Martin Pawelczyk: Campus, Ilias (online ab 14.10.19, 8:30 Uhr)

Das Seminar bietet eine Einführung in das noch junge Forschungsgebiet der algorithmischen Fairness. Im Seminar werden wir aktuelle Entwicklungen besprechen und auf die neusten Forschungsergebnisse eingehen. Für die Teilnahme am Seminar  werden grundlegende Kentnisse aus den Bereichen der Statistik und des Maschinellen Lernens vorausgesetzt. Außerdem werden gute Programmierkentnisse alsauch ein grundsätzliches Interesse an sozialen Themen angenommen, da sich der Answendungsbereich des Forschungsfelds inbesondere an sozial-technischen Problemen orientiert (z.B. Credit Scoring, College Admissions, Recidivism Prediction).

Lehrplan

• Fairness Definitions
• Causality Based Fairness Defintions
• Impossibility Theorems
• Learning Fair Representations (pre-processing)
• Post-processing methods
• ML algorithms subject to fairness constraints

Informationen

Dozent: Gjergji Kasneci; Organisation: Martin Pawelczyk

Die Vorlesung bietet eine Einführung in fortgeschrittene Methoden zum Finden von Mustern in großen Datensätzen. Es werden verschiedene Techniken aus den Bereichen Maschinelles Lernen, Statistik, Wahrscheinlichkeitstheorie und Informationssysteme behandelt.

Lehrplan

• Grundlagen der Wahrscheinlichkeitstheorie und Statistik
• Grundlagen der Informationstheorie
• Daten Vorverarbeitung
• Daten-Indizierung für effiziente Ähnlichkeitssuche (Similarity Search)
• Überwachtes (supervised) Lernen: Einführung in Klassifikationsverfahren
• Lineare und Nicht-Lineare Klassifikation
• Regression

Informationen

Dozent: Gjergji Kasneci; Übungsleitung: Johannes Haug, Vadim Borisov; ILIAS; Campus

In many practical scenarios, the abundance of data provided by many information sources makes it probable to find that different sources provide partially or completely different values for the same data item. And sometimes the information sources do not provide any values at all. In this seminar, we discuss probabilistic techniques for improving the quality of the generated data by inferring the correct values for the provided data items and the reliability of the information sources.

Information

Lecturer: Gjergji Kasneci