Portfolio item number 1
Short description of portfolio item number 1
Posts by Collection
portfolio
Portfolio item number 2
Short description of portfolio item number 2
publications
Irregular layout for a satellite’s interferometric array
Published in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021
We propose a solver to optimize the antenna placement on satellite-mounted interferometric synthetic-aperture instruments.
Recommended citation: Krzakala, Paul and Assouel, Amine, et al. (2021). Irregular layout for a satellite’s interferometric array. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 14, 9408-9423..
Download Paper
Any2Graph: Deep End-To-End Supervised Graph Prediction With An Optimal Transport Loss
Published in NeurIPS 2024, 2024
A fully differentiable framework for Supervised Graph Prediction.
Recommended citation: Krzakala, P., Yang, J., Flamary, R., d'Alché-Buc, F., Laclau, C., & Labeau, M. (2024). Any2Graph: Deep End-To-End Supervised Graph Prediction With An Optimal Transport Loss. Advances in Neural Information Processing Systems, 37, 101552-101588.
Download Paper
The Quest for the GRAph Level autoEncoder (GRALE)
Published in arXiv preprint, 2025
Self-supervised graph representation learning using a trainable graph matching module for computing the reconstruction loss.
Recommended citation: Krzakala, P., Melo, G., Laclau, C., d'Alché-Buc, F., & Flamary, R. (2025). The Quest for the GRAph Level autoEncoder (GRALE). arXiv preprint arXiv:2505.22109.
Download Paper
talks
An introduction to Machine Learning.
Published:
Demystification, foundations and state of the art.
Download Download Slides
Adversarial Examples, an overview.
Published:
The intruiging properties of Adversarial Examples, the never ending race between attacks and defences and the most popular explanations.
Download Download Slides
Any2Graph, A framework for deep Supervised Graph Prediction
Published:
We introduce Any2Graph, a novel framework for deep Supervised Graph Prediction.
Download Download Slides
Training Neural Networks to Predict Graphs. Who is afraid of the big bad NP-hardness ?
Published:
Graph learning is a rapidly growing area of machine learning, with most advances focused on problems where graphs are the input. Yet, many important tasks involve predicting graphs as the output. This setting poses a fundamental challenge: evaluating predictions requires solving the graph matching problem, an NP-hard task. In this talk I will: (a) review existing strategies to circumvent this obstacle, such as graph canonization and generative approaches; (b) demonstrate that, in a data-driven setting, the hardness barrier can effectively disappear as, for a given graph distribution, it is possible to learn a scalable solver; and (c) showcase how this strategy can be used to train a graph-level autoencoder i.e. a model that learns to encode (and decode) entire graphs into a vector space.
Download Download Slides
teaching
Machine Learning
Teaching Assistant, Télécom Paris, Institut Polytechnique, 1900
SD-TSIA210
Structured Data: Learning and Prediction
Teaching Assistant, Télécom Paris, Institut Polytechnique, 1900
MAP670I (DATA917)