The State-Of-the-Art in

Explanation Methods

for Two-Dimensional Embeddings of Data

A tutorial on explainability methods for dimensionality-reduction-based embeddings

August 15/16/17 (TBD), IJCAI-ECAI 2026, Bremen, Germany

Overview

Dimensionality reduction methods are widely used to create two-dimensional views of high-dimensional data, but it is not so straightforward how to make sense of apparent patterns such as clusters or paths. This tutorial offers a structured guide to which explainability techniques exist that link these patterns back to the original features, ranging from feature-importance and counterfactual approaches to interactive annotation of low-dimensional embeddings.

It mainly contains:

Challenges of interpreting DR embeddings, with examples from life sciences and text analysis; defining the survey scope and methodology;
A newly developed taxonomy, organized by the analysis task and target, the type of explanation produced, the visualization strategy, and the ways these methods are evaluated;
Guidelines on how to select methods based on task, data type, DR technique, desired outputs;

We are still finalizing the contents and material of the tutorial. Slides and material will be posted here before the event.

Key details

Time: August 15/16/17 (to be determined by the conference organizers)
Location: IJCAI-ECAI 2026, Bremen, Germany
Presenters: Edith Heiter, Fuyin Lai, Jefrey Lijffijt

Tentative schedule

The survey underlying this tutorial covers 40 explainability methods for two-dimensional embeddings, which are arranged in a multi-faceted taxonomy. In the single 1 hour 45-minute slot, we will guide attendees through a newly developed taxonomy, explain several representative methods in-depth, and discuss practical guidelines and open challenges. By the end of the tutorial, you will have a clear view on what methods exist, providing you with practical guidance on choosing a method for a given use case, and an understanding of the open challenges.

0 - 10 min

Introduction & motivation

10 - 85 min

Taxonomy and overview of various aspects

- Taxonomy overview (5 min)

- Analysis tasks and explanation target (10 min)

- Explanation output types (30 min)

- Visualization strategies (10 min)

- Evaluation (10 min)

85 - 95 min

Practical guidelines

95 - 105 min

Open problems & future directions

105 - 115 min

Q & A

Speakers

Edith Heiter

Edith Heiter is a PhD student at Ghent University within the Artificial Intelligence & Data Analytics Research Group at IDLab. She holds both a B.Sc. (2016) and an M.Sc. (2020) in Computer Science from Saarland University, Germany. Under the supervision of Prof. Jefrey Lijffijt and Prof. Tijl De Bie, her research focuses on the intersection of dimensionality reduction, data visualization, and explainability. She has served as a teaching assistant for several years for courses covering Big Data Technologies and Data Visualization for and with AI.

Fuyin Lai

Fuyin Lai is a PhD student at Ghent University’s IDLab, Department of Electronics and Information Systems. He earned his B.Sc. in Computer Science from Hefei University of Technology and an M.Sc. in Mathematics from the Harbin Institute of Technology. Supervised by Prof. Jefrey Lijffijt and Prof. Tijl De Bie, his research interests include data exploration, dimensionality reduction methods, and data visualization. Since 2024, he has been a teaching assistant for master-level courses on Big Data and AI-driven data visualization.

Jefrey Lijffijt

Jefrey Lijffijt is a Professor of Data Science, Knowledge Discovery, and Visual Analytics at Ghent University, Belgium. He received his D.Sc.(Tech.) from Aalto University in 2013 and held research positions at the University of Bristol and as an FWO Marie Skłodowska-Curie Fellow before his current appointment. His expertise spans statistical modeling, machine learning, and visual analytics. An experienced educator and speaker, he teaches data science and machine learning courses to both university students and professionals, and has lectured several tutorials before at the ECML-PKDD conference.

Page updated

Google Sites

Report abuse