PhD Position Scientific Machine Learning, Toward Scientific Foundation Models

PhD Position Scientific Machine Learning, Toward Scientific Foundation Models

Passionate about advancing foundation models for science? Join our PhD project at TU Delft!

We invite applications for a fully funded PhD position in the area of Scientific Machine Learning (SciML), which integrates data-driven machine learning techniques with established scientific knowledge, such as physical laws, differential equations, and domain-specific constraints, to model, simulate, and understand complex systems. The project will explore modern SciML methods, such as physics-informed neural networks, neural operators (e.g., Fourier Neural Operators) and hybrid physics-ML approaches.

These models are expected to play a significant role in scientific domains and critical applications such as climate and geoscience, as well as the energy sector (for example, subsurface modeling, seismic inversion, climate prediction, renewable energy forecasting, and power grid optimization).

Building on this, the project focuses on the definition, development, and analysis of scientific foundation models: large-scale, generalizable models trained across diverse scientific datasets that aim to capture the underlying principles of physical systems and can be adapted to a wide range of tasks. Within this broad theme, the PhD project can take several possible directions, including:

• Developing scientific foundation models for inverse problems—moving beyond forward simulation toward tasks such as inferring hidden physical parameters, reconstructing unknown states, or identifying governing mechanisms from indirect or partial observations.
• Developing uncertainty-aware methods that can identify unreliable predictions and indicate where additional data would be most valuable.
• Studying how such foundation models generalize across related but distinct physical settings, such as changes in boundary conditions, geometries, parameters, or forcing terms.
• Exploring their potential to accelerate or complement conventional numerical simulations.

The successful candidate will join a multidisciplinary research environment at the intersection of machine learning, applied physics, and domain sciences.

• MSc degree in computer science, artificial intelligence, applied mathematics, applied physics, data science, or a closely related field.
• Good theoretical understanding of the fundamentals of machine and deep learning, with a strong interest in methodological development rather than only implementation and application.
• Basic knowledge and a keen interest in physical problems (especially inverse problems) and scientific applications.
• Strong programming skills (preferably Python).
• Ability to work independently (taking initiative, being organized) and to collaborate effectively.
• Strong ability in research communication and interpersonal communication.

English proficiency at a suitable level is required to participate in English-taught Doctoral Education courses and to write scientific articles and a final thesis.

Doctoral candidates will be offered a 4-year period of employment, implemented through two contracts: an initial 1.5-year contract with an official go/no‑go progress assessment within 15 months, followed by an additional 2.5-year contract assuming satisfactory performance.

Salary and benefits are in accordance with the Collective Labour Agreement for Dutch Universities, ranging from €3,059 to €3,881 gross per month from the first to the fourth year, based on a full‑time contract (38 hours), plus 8 % holiday allowance and an end‑of‑year bonus of 8.3 %. Additional benefits include a customizable compensation package, discounts on health insurance, a monthly work‑costs contribution, and flexible work schedules.

Contact information:
For more information about this vacancy or the selection procedure, please contact Dr. Jing Sun, via jing.sun.