HI I'M DESPOINA TOUSKA

Despoina Touska
Machine Learning Scientist
Status
Open to work
Contact
CV IN @
Amsterdam, NL
My Experience
Teaching Assistant at UvA
2025 - Netherlands - Part time
Research Intern at ASML
2024/5 - Netherlands - Full time
Machine Learning Scientist at UvA
2024 - Netherlands - Full time
Machine Learning Scientist at CERTH
2020/3 - Greece - Full time
Research Intern at ISEP
2019/20 - Portugal - Full time
Research Intern at CERTH
2018 - Greece - Full time
About Me

I am a Machine Learning Researcher and Engineer with a strong background in computer engineering and deep learning. I hold an MSc in Artificial Intelligence, awarded cum laude, from the University of Amsterdam, and an MEng in Electrical and Computer Engineering from Aristotle University of Thessaloniki.

My work focuses on computer vision, representation learning, generative AI, and efficient machine learning methods. Beyond conventional AI, I am interested in physics-inspired and neuroscience-inspired AI, particularly in how principles from the natural sciences can inform learning systems and how AI can drive scientific discovery.

Selected Publications

OrthoRF: Exploring Orthogonality in Object-Centric Representations

Touska, D., Auer, B., Onose, A., Kasarla, T., Pérez Rey, L., Lipp, M., Amitonova, L., Oswald, M., Cerfontaine, P.

ICLR 2026

Project Page ↗

Graph-based data association in multiple object tracking: A survey

Touska, D., Gkountakos, K., Ioannidis, K., Tsikrika, T., Vrochidis, S., Kompatsiaris, I.

MMM 2023

Spatio-temporal activity detection and recognition in untrimmed surveillance videos

Gkountakos, K., Touska, D., Ioannidis, K., Tsikrika, T., Vrochidis, S., Kompatsiaris, I.

ICMR 2021

Detecting tampered videos with multimedia forensics and deep learning

Zampoglou, M., Markatopoulou, F., Mercier, G., Touska, D., Apostolidis, E., Papadopoulos, S., Kompatsiaris, I.

MMM 2019

ITI-CERTH participation in ActEV and AVS tracks of TRECVID 2021

Gkountakos, K., Galanopoulos, D., Touska, D., Ioannidis, K., Vrochidis, S., Mezaris, V., Kompatsiaris, I.

NIST TRECVID Workshop 2021

Blog

Orthogonality as an Inductive Bias for Object-Centric Learning

April 2026

Notes on rotating features, object-centric learning, and how orthogonal latent directions can help separate visual entities in overlapping scenes and recover occluded object parts.

CV Neuroscience Read ↗

Understanding Barlow Twins: Similarity and Redundancy Reduction in SSL

May 2026

Barlow Twins is a self-supervised method that encourages two augmented views of the same image to produce similar embeddings, while reducing redundancy between feature dimensions.

SSL Read ↗