Research

My main research interests are on unsupervised (or self-supervised) machine learning, particularly with geometrically-inspired representations. I investigate how the geometry of representations learned by different models can be either enforced or understood through the formalism of Riemannian manifolds.

During my PhD, I worked on one of the first temporal graph neural networks for recommender systems. In generative modeling, I investigated how to improve the generation of data via interpolations in the base space of normalising flows.

Prior to my PhD research, my research interests focused on data visualization, with an emphasis on making high-dimensional data understandable through interactive 2D scatterplots.

Teaching

I have been regularly teaching courses every semester since 2020. I mostly teach for graduate (MSc. level) courses, but in the past I have worked as a teaching assistant for BSc. courses.

When opportunity allows it, I also try to come up with alternative ways to help students understand some concepts. Particularly for a Deep Learning course, I have designed a visualization tool for convolutions. I would be happy to hear if you found it useful or have feedback.

Publications

• Towards General Geometries for Embedding Knowledge Graphs
  • S. G. Fadel*, T. Paulsen*, S. Mair
  • ICML/ELLIS Workshop on Geometry-grounded Representation Learning and Generative Modeling, 2024
  • HTML
• Self-Supervised Siamese Autoencoders
  • F. Baier, S. Mair, S. G. Fadel
  • Advances in Intelligent Data Analysis XXII, 2024
  • arXivDOI
• Exploring the Poincaré Ellipsis
  • S. G. Fadel*, T. Paulsen*, U. Brefeld
  • International Workshop on Mining and Learning with Graphs (ECML/PKDD), 2023
• Studying the Propagation of Information in VAE Decoders
  • Y. Rudolph, S. G. Fadel, S. Mair, U. Brefeld
  • Northern Lights Deep Learning Workshop, 2022
• Principled Interpolation in Normalizing Flows
  • S. G. Fadel*, S. Mair*, R. da S. Torres, U. Brefeld
  • Proceedings of the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML/PKDD), 2021
  • slidesposterarXivDOI
• Contextual Movement Models Based on Normalizing Flows
  • S. G. Fadel*, S. Mair*, R. da S. Torres, U. Brefeld
  • AStA Advances in Statistical Analysis, Special Issue on Statistics in Sports, 2021
  • DOI
• Efficient Normalizing Flows to Polytopes
  • S. G. Fadel, S. Mair, R. da Silva Torres, U. Brefeld
  • Northern Lights Deep Learning Workshop, 2020
• An Appropriate Prior Distribution for Interpolating Latent Samples in Flow-based Generative Models (abstract)
  • S. G. Fadel, S. Mair, R. da Silva Torres, U. Brefeld
  • Northern Lights Deep Learning Workshop, 2020
• Neural Relational Inference for Disaster Multimedia Retrieval
  • S. G. Fadel, R. da S. Torres
  • Multimedia Tools and Applications, 2020
  • DOI
• Link Prediction in Dynamic Graphs for Recommendation
  • S. G. Fadel, R. da S. Torres
  • NeurIPS Workshop on Relational Representation Learning, 2018
  • arXiv
• Graph-based Early Fusion for Flood Detection
  • R. de O. Werneck, Í. C. Dourado, S. G. Fadel, S. Tabbone, R. da S. Torres
  • IEEE International Conference on Image Processing (ICIP), 2018
  • DOI
• UPDis: A User-Assisted Projection Technique for Distance Information
  • T. A. T. T. Neves, S. G. Fadel, G. M. Hilasaca, F. M. Fatore, F. V. Paulovich
  • Information Visualization, 2018
  • DOI
• Exploiting ConvNet Diversity for Flooding Identification
  • K. Nogueira, S. G. Fadel, Í. C. Dourado, R. de O. Werneck, J. A. V. Muñoz, O. A. B. Penatti, R. T. Calumby, L. T. Li, J. A. dos Santos, R. da S. Torres
  • IEEE Geoscience and Remote Sensing Letters, 2018
  • DOI
• Visualizing the Hidden Activity of Artificial Neural Networks
  • P. E. Rauber, S. G. Fadel, A. X. Falcão, A. C. Telea
  • IEEE Transactions on Visualization and Computer Graphics, 2016
  • DOI
• LoCH: A Neighborhood-based Multidimensional Projection Technique for High-Dimensional Sparse Spaces
  • S. G. Fadel, F. M. Fatore, F. S. L. G. Duarte, F. V. Paulovich
  • Neurocomputing, 2015
  • DOI
• On the Effectiveness of User Manipulation in Multidimensional Projections
  • S. G. Fadel, F. V. Paulovich
  • Conference on Graphics, Patterns and Images, XXVIII; Workshop on Visual Analytics, Information Visualization and Scientific Visualization, 2015
• Nmap: A Novel Neighborhood Preservation Space-filling Algorithm
  • F. S. L. G. Duarte, F. Sikansi, F. M. Fatore, S. G. Fadel, F. V. Paulovich
  • IEEE Transactions on Visualization and Computer Graphics, 2014
  • DOI

Background

Postdoc researcher (2024─present)

• Section for Cognitive Systems with Søren Hauberg
• DTU Compute, Technical University of Denmark

Postdoc researcher (2023─2024)

• Division of Statistics and Machine Learning with Fredrik Lindsten
• Department of Computer and Information Science, Linköping University

Postdoc researcher (2021─2023)

• Machine Learning Group with Ulf Brefeld
• Institute of Information Systems, Leuphana University

Visiting researcher (2019─2020, 15 months)

• Machine Learning Group
• Supervised by Ulf Brefeld

Ph.D. in Computer Science (2016─2021)

• "Learning in non-Euclidean Domains"
• Institute of Computing, University of Campinas
• Supervised by Ricardo da S. Torres

Visiting researcher (2015─2016, 3 months)

• Scientific Visualization and Computer Graphics
• Supervised by Alex Telea

M.Sc. in Computer Science (2014─2016)

• "Understanding Interactive Multidimensional Projections"
• Institute of Mathematical and Computer Sciences, University of São Paulo
• Supervised by Fernando Paulovich

B.Sc. in Computer Science (2010─2013)

• "Multidimensional Projections based on Convex Hulls"
• Institute of Mathematical and Computer Sciences, University of São Paulo