Scribd
Upload
9 views3 pages

Ph.D. Position On Uncertainty-Aware Geometric Graph Neural Networks and Applications

A Ph.D. position is available at Université Paris-Saclay focusing on uncertainty-aware geometric graph neural networks (GNNs) for applications in materials modeling. The research aims to develop models that balance predictive performance and computational efficiency while incorporating uncertainty estimation mechanisms. Candidates should have a Master's degree in a relevant field, strong programming skills, and an interest in machine learning applications in biomedical contexts.

Uploaded by

Kaito
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
9 views3 pages

Ph.D. Position On Uncertainty-Aware Geometric Graph Neural Networks and Applications

A Ph.D. position is available at Université Paris-Saclay focusing on uncertainty-aware geometric graph neural networks (GNNs) for applications in materials modeling. The research aims to develop models that balance predictive performance and computational efficiency while incorporating uncertainty estimation mechanisms. Candidates should have a Master's degree in a relevant field, strong programming skills, and an interest in machine learning applications in biomedical contexts.

Uploaded by

Kaito
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Ph.D.

Position on Uncertainty-aware Geometric Graph


Neural Networks and Applications
▶ Keywords: Graph representation learning, geometric deep learning, graph neural networks,
uncertainty quantification, molecular systems
▶ Research group: Université Paris-Saclay CentraleSupélec, Centre de la Vision Numérique,
Inria Saclay–OPIS team
▶ Supervisor: Fragkiskos Malliaros ([email protected])
▶ Conditions: This Ph.D. topic is participating in the DeMythif.AI program competition of
Université Paris-Saclay. It is reserved for international students who have spent less than 12
months in France in the last 3 years. The candidates will be evaluated by a jury that will select
20 Ph.D. theses to start in Fall 2025. The successful candidates will be fully funded for 3 years,
have access to specific scientific and non-scientific training, and be fully part of the Université
Paris-Saclay AI community.

Context and objectives

Graph Neural Networks (GNNs) (e.g., [1,2]) currently constitute state-of-the-art models for solving
prediction tasks on graphs. Through the flexible formulation of the message passing mechanism,
GNNs can learn informative latent representations of graph entities at different resolution levels
(e.g., node-, edge-, graph-level). In many practical applications in molecular and chemical systems,
the nodes of the graph have associated geometric attributes (e.g., coordinates, velocities) related to
their position in the 3D space. In this context, geometric graphs represent the interaction of atoms in
the 3D space, encapsulating a range of physical symmetries such as rotations and translations [3].
Existing GNN models often overlook this aspect, rendering them ill-suited for prediction tasks on
geometric graphs. Recently, Geometric GNN architectures tailored to respect physical symmetries
have emerged as flexible models on atomic systems such as small molecules, proteins, and mate-
rials [4, 5] (please refer to our recent survey article [6] for a detailed exploration of the field). Nev-
ertheless, such models often have to balance expressiveness and computational efficiency due to
the requirement to learn invariant or equivariant functions. Besides, they lack rigorous uncertainty
estimates that will allow practitioners to assess the reliability of the predictions.
The goal of this Ph.D. position is to develop uncertainty-aware geometric GNN models that can
be leveraged in real-world prediction problems, primarily in materials modeling [7]. The objectives
of the position are the following:
• Objective 1. First, we aim to develop geometric GNNs that balance predictive performance
and scalability. Following our recent work [8], we will study methodologies to enforce sym-
metries through appropriate data augmentation strategies rather than directly in the model
architecture.
• Objective 2. Then, we will investigate uncertainty estimation and quantification mechanisms
in geometric GNNs. This includes Bayesian formulations of GNNs in which the message pass-
ing mechanism propagates uncertainty estimates of nodes beyond feature representations.
Besides, we are interested in studying a conformal prediction framework for model-agnostic
uncertainty estimates [9].

Throughout the Ph.D. project, we will investigate practical applications of the proposed models
in complex molecular systems arising in materials modeling. This will be achieved in collabora-
tion with experts in the broader context of leveraging machine learning techniques to analyze the
properties of materials to address energy challenges posed by climate change [10].

Candidate profile

In addition to the conditions specified in the DeMythif.AI competition described above, the candi-
date student should have:
• A Master’s (M2) degree in Engineering, Computer Science, Applied Mathematics, or a re-
lated discipline, with a strong background in machine learning, graph theory, and probability
theory.
• The candidate should be interested in exploring machine learning methods on graph-structured
data and their real-world biomedical applications.
• Strong programming skills in Python (including PyTorch).
• Good communication skills.

Envisioned collaborations

In the context of the Ph.D. thesis, there will be a collaboration with David Rolnick, Assistant Pro-
fessor at McGill University and Mila – Quebec AI Institute, and chair of the Climate Change AI
initiative. Through this collaboration, the application part of the uncertainty-aware geometric GNN
models will be developed in the context of AI-driven materials modeling in climate change.

Team and location

CentraleSupélec is a premier engineering school in France and a constituent member of Université


Paris-Saclay (consistently ranked among the best universities worldwide). CentraleSupélec is lo-
cated on the outskirts of Paris (around 45 minutes by train from the center of Paris) at the center of
the Paris-Saclay cluster—a fast-growing research and industrial ecosystem. The Ph.D. student will
be integrated within the Centre for Visual Computing (CVN), a joint laboratory between Centrale-
Supélec and Inria Saclay.

2
References
[1] T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” in ICLR,
2017.
[2] P. Veličković, G. Cucurull, A. Casanova, A. Romero, P. Liò, and Y. Bengio, “Graph attention networks,”
in ICLR, 2018.
[3] X. Zhang et al., “Artificial intelligence for science in quantum, atomistic, and continuum systems,” CoRR,
2023.
[4] Y. Liu, L. Wang, M. Liu, Y. Lin, X. Zhang, B. Oztekin, and S. Ji, “Spherical message passing for 3d
molecular graphs,” in ICLR, 2022.
[5] Z. Zhang, M. Xu, A. R. Jamasb, V. Chenthamarakshan, A. Lozano, P. Das, and J. Tang, “Protein repre-
sentation learning by geometric structure pretraining,” in ICLR, 2023.
[6] A. Duval, S. V. Mathis, C. K. Joshi, V. Schmidt, S. Miret, F. D. Malliaros, T. Cohen, P. Lio, Y. Bengio, and
M. Bronstein, “A hitchhiker’s guide to geometric gnns for 3d atomic systems,” CoRR, 2024.
[7] C. L. Zitnick et al., “An introduction to electrocatalyst design using machine learning for renewable
energy storage,” CoRR, 2020.
[8] A. Duval, V. Schmidt, A. H. Garcia, S. Miret, F. D. Malliaros, Y. Bengio, and D. Rolnick, “FAENet: Frame
averaging equivariant gnn for materials modeling,” in ICML, 2023.
[9] K. Huang, Y. Jin, E. J. Candès, and J. Leskovec, “Uncertainty quantification over graph with conformal-
ized graph neural networks,” CoRR, 2023.
[10] L. Kaack, P. Donti, E. Strubell, G. Kamiya, F. Creutzig, and D. Rolnick, “Aligning artificial intelligence
with climate change mitigation,” Nature Climate Change, vol. 12, 06 2022.

You might also like