Nikolaos Dimitriadis

PhD student

EPFL

Hello! I am Nikos

I am a PhD student in Computer Science at École Polytechnique Fédérale de Lausanne (EPFL), where I am advised by François Fleuret and Pascal Frossard. My research interests revolve around model merging, multi-task and continual learning. From December 2023 to December 2024, I interned in Google DeepMind in Zurich (full-time for 3 months and part-time for the rest of the year), where I worked on text-to-image generation.

Before coming to Switzerland, I completed my undergraduate studies in Electrical and Computer Engineering at the National Technical University of Athens in Greece. I conducted my thesis (available here in Greek) at the CVSP lab under the supervision of Petros Maragos. The focus lied on using tropical geometry to analyze Morphological Neural Networks, studying the sparsity of their representations compared to their linear counterparts, their ability to enforce shape constraints such as monotonicity, and extending a training algorithm based on Difference-of-Convex Programming to multiclass problems.

I am also an avid classical guitar player! I love playing Baroque and romantic pieces, such as compositions by Agustín Barrios Mangoré. Check out this beautiful performance!

Download my resumé .

Interests

Model Merging
Multi-Task Learning
Continual Learning

Education

PhD in Computer Science

École Polytechnique Fédérale de Lausanne
MEng in Electrical Engineering and Computer Science, 2020

National Technical University of Athens

Featured Publications

Ke Wang, Nikolaos Dimitriadis, Guillermo Ortiz-Jimenez, Francois Fleuret, Pascal Frossard

January, 2024 International Conference on Machine Learning (ICML)

Localizing Task Information for Improved Model Merging and Compression

Nikolaos Dimitriadis, Pascal Frossard, Francois Fleuret

July, 2023 International Conference on Machine Learning (ICML)

Pareto Manifold Learning: Tackling Multiple Tasks via Ensembles of Single-Task Models

We present a novel approach to multi-task learning, that utilizes weight ensembles to produce a continuous Pareto Front in a single training run.

Publications

Quickly discover relevant content by filtering publications.

Ke Wang, Nikolaos Dimitriadis, Alessandro Favero, Guillermo Ortiz-Jimenez, Francois Fleuret, Pascal Frossard (2025). LiNeS: Post-training layer scaling prevents forgetting and enhances model merging. International Conference on Learning Representations (ICLR).

Cite PDF GitHub Repository Website

Nikolaos Dimitriadis, Pascal Frossard, Francois Fleuret (2025). Pareto Low-Rank Adapters: Efficient Multi-Task Learning with Preferences. International Conference on Learning Representations (ICLR).

Cite PDF

Ke Wang, Nikolaos Dimitriadis, Guillermo Ortiz-Jimenez, Francois Fleuret, Pascal Frossard (2024). Localizing Task Information for Improved Model Merging and Compression. International Conference on Machine Learning (ICML).

Cite PDF Website GitHub repository

Kyle Matoba, Nikolaos Dimitriadis, Francois Fleuret (2023). Benefits of Max Pooling in Neural Networks: Theoretical and Experimental Evidence. Transactions on Machine Learning Research.

PDF Cite URL

Nikolaos Dimitriadis, Pascal Frossard, Francois Fleuret (2023). Pareto Manifold Learning: Tackling Multiple Tasks via Ensembles of Single-Task Models. International Conference on Machine Learning (ICML).

PDF Cite Poster Slides GitHub repository

Nikolaos Dimitriadis, Francois Fleuret, Pascal Frossard (2023). SequeL: A Continual Learning Library in PyTorch and JAX. CVPR Workshop on Continual Learning 2023 (non-archival track).

PDF Cite Website GitHub repository

Ahmet Caner Yüzügüler, Nikolaos Dimitriadis, Pascal Frossard (2023). Flexible Channel Dimensions for Differentiable Architecture Search. arXiv (Preprint).

PDF Cite arXiv

Ahmet Caner Yüzügüler, Nikolaos Dimitriadis, Pascal Frossard (2022). U-Boost NAS: Utilization-boosted Differentiable Neural Architecture Search. European Conference on Computer Vision (ECCV).

PDF Cite Poster Slides Video GitHub repository

Nikolaos Dimitriadis, Petros Maragos (2021). Advances in Morphological Neural Networks: Training, Pruning and Enforcing Shape Constraints. ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

PDF Cite Poster Slides Video

Nikolaos Dimitriadis, Petros Maragos (2020). Advances in the Training, Pruning and Enforcement of Shape Constraints of Morphological Neural Networks Using Tropical Algebra. arXiv (Preprint).

PDF Cite arXiv

Contact

nikolaos.dimitriadis[at]epfl.ch