ESA Graduate Trainee in Scientific Deep Learning

Our team and mission

The Technology Department is responsible for the technology strategy, the research and technology development
programmes, the education programme and the directorates communication activities.

In particular, this includes, together with all relevant directorates:

• developing and implementing ESA’s technology strategy;
• organising studies, research and developments to provide an integrated, continuous technology development path
  from TRL1 to TRL9 according to strategic and programmatic needs, available competences and resources;
• coordinating and harmonising technology developments with ESA’s application and programme specific technology
  development programmes, European and national technology development programmes;
• developing and implementing a resource and competence plan for conducting R&D activities, together with the
  Management Support Office and the other departments;
• preparing future missions and their technologies through early phase studies, system analyses, feasibility
  assessments, and establishing mission baselines for DG, Directorate, and Member State decisions;
• developing together with the TEC business partners, Senior Technical Authorities and System engineers effective
  R&D processes addressing user and programme needs;
• liaising with D/CIC on commercialisation and competitiveness aspects of R&D activities, and ensure the alignment
  of ESA's technology strategy with the strategies in these domains;
• liaising with D/OPS on ground system R&D activities;
• integrating relevant Education activities into the R&D management processes;
• communicating the value of ESA’s technical competence, infrastructure and facilities;
• coordinating in close cooperation with corporate communication at ESA, TEC communication activities;
  managing TEC internal communication activities.

This EGT will take place within ESA's Advanced Concepts Team (ACT). ESA’s Advanced Concepts Team (ACT) monitors, performs, and promotes cutting-edge multidisciplinary research for space. It explores innovative approaches to space-related R&D, including competitions, prizes, and games, as well as research aimed
at fostering disruptive innovation. The team develops an expert network within academia and provides rapid first-look
analyses of challenges, opportunities, and problems.
The ACT collaborates with universities and research centers, focusing on advanced topics of strategic relevance to the space
sector while experimenting with novel teamwork methods. To achieve its objectives, the ACT fosters a dynamic,
multidisciplinary research environment where early-career researchers—spanning postdoctoral and postgraduate levels in
science and engineering—contribute to the development of emerging technologies and innovative concepts.

You are encouraged to visit the ACT website: www.esa.int/act as well as the ESA website: http://www.esa.int

Field(s) of activity/research for the traineeship

You will carry out most of your activities in the field of scientific deep learning, with a strong focus on geometric deep
learning and other advanced machine learning techniques designed to learn from structured, relational, and physically
grounded data. Scientific deep learning aims to integrate data-driven methods with domain knowledge, physical constraints,
and mathematical structure in order to build models that are robust, interpretable, and learn in data-scarce settings.
Geometric deep learning [1] extends classical neural networks to non-Euclidean domains such as graphs, meshes, manifolds,
and spheres. These methods are particularly well suited to problems where the underlying data exhibit symmetries,
conservation laws, or relational structure. You will explore how such techniques can be used to model complex systems,
learn representations of physical processes, and enable more reliable inference and prediction in scientific and engineering
contexts such as gravity inversion [2], optimal control [3,4], advanced materials [5].

You are encouraged to propose your own research ideas within this broad scope. Possible activities include, but are not
limited to:

• learning on sphere, graphs, meshes, and manifolds for modelling physical systems and relational data;
• equivariant and invariant neural architectures that respect geometric symmetries;
• hybrid approaches combining deep learning with numerical simulation and optimisation;
• representation learning for high-dimensional, sparse, or multi-modal scientific data;
• physics-informed and constraint-aware neural networks that incorporate prior knowledge and governing equations;
• uncertainty-aware and probabilistic deep learning methods for scientific applications;
• development of open-source research software and reproducible machine learning pipelines.

Throughout the traineeship, you will strengthen your understanding of modern machine learning theory while gaining
hands-on experience with applications on dynamical systems. You will learn how to translate scientific questions into
machine learning problems, design appropriate architectures exploiting symmetries in the data, evaluate models critically,
and communicate results clearly to both technical and non-technical audiences.

As a member of the Advanced Concepts Team (ACT), you will contribute to the development and evaluation of new space
technologies and concepts. You will collaborate with experts from diverse fields, including artificial intelligence, computer
science, fundamental physics, and mission analysis. Depending on your background and interests, your work may include
various initiatives, such as competitions organized via the ESA’s ACT optimize platform (https://optimize.esa.int) and/or
studies conducted under ESA’s Ariadna scheme, and you will help disseminate research findings within ESA and to external
audiences.

Finally, you will monitor—and if feasible, contribute to—ESA's Discovery and Preparation campaigns by refining early study
definitions and possibly participating in select activities.

.

References:
[1] Bronstein, M. M., Bruna, J., Cohen, T., & Veličković, P. (2021). Geometric deep learning: Grids, groups, graphs, geodesics,
and gauges. arXiv preprint arXiv:2104.13478.
[2] Izzo, Dario, and Pablo Gómez. "Geodesy of irregular small bodies via neural density fields." Communications
Engineering 1.1 (2022)
[3] Izzo, Dario, et al. "Optimality principles in spacecraft neural guidance and control." Science Robotics 9.91 (2024):
eadi6421,
[4] Izzo, Dario, et al. "High-order expansion of Neural Ordinary Differential Equations flows." arXiv preprint arXiv:2504.08769
(2025).
[5] Dold, Dominik, and Derek Aranguren van Egmond. "Differentiable graph-structured models for inverse design of lattice
materials." Cell Reports Physical Science 4.10 (2023).

Technical competencies

Behavioural competencies

Result Orientation

Operational Efficiency

Fostering Cooperation

Relationship Management

Continuous Improvement

Forward Thinking

For more information, please refer to ESA Core Behavioural Competencies guidebook

Education

You should have just completed, or be in the final year of your master’ s degree in . [insert only the discipline] [all other requirement goes under Additional requirements]

Additional requirements

You should have good interpersonal and communication skills and should be able to work in a multicultural environment, both independently and as part of a team. Previous experience of working in international teams can be considered an asset. Your motivation, overall professional perspective and career goals will also be explored during the later stages of the selection process. 

You should also have:

Basic knowledge of machine learning pipelines and best practices.
Hands-on experience with Python and/or C++, and state-of-the-art machine learning tools such as PyTorch, TensorFlow, or JAX.
A solid understanding of core mathematical concepts such as linear algebra, probability theory, optimization, and statistical
learning theory.
Strong inclination toward the theoretical foundations of machine learning and deep learning.