Are you a master or PhD-student and looking for an exciting fully funded research internship this summer? Do you want to apply your skills to real-world industrial challenges while working in an international team?
The Graduate-Level Research in Industrial Projects for Students (G-RIPS) Berlin is your opportunity to collaborate on cutting-edge industry-sponsored research at the Research Campus MODAL located at the Zuse Institute Berlin (ZIB). Over eight weeks, you’ll work in cross-cultural teams tackling high-impact problems in both analytical and computational research. (See list of projects below.)
The program will run from June 22 to August 14, 2026.
G-RIPS is a joint activity of ZIB (Research Campus MODAL) and IPAM (Institute for Pure & Applied Mathematics (at) UCLA).
Tasks
As a G-RIPS participant, you’ll:
✔ Work in international, cross-cultural teams with students from the U.S. and Europe.
✔ Solve challenging industry-sponsored research problems, gaining valuable real-world experience. (Please see the list of this years projects below. You can only work in ONE project during the internship.)
✔ Conduct both analytical and computational research, applying mathematical and data-driven approaches.
Requirements
Eligible applicants include master and PhD students from the areas of mathematics, computer or natural sciences (and related fields) who are currently enrolled at a European university. Due to the large number of applications we typically receive, we do not accept applications from previous RIPS or G-RIPS students.
Please submit the following documents
-
motivation (cover) letter (please state your preferred project),
-
resume/CV,
-
an academic record or transcript (can be unofficial).
Apply now and secure your spot in this fully funded international research internship!
Benefits
Costs for traveling and accommodation in Berlin are covered.
Here is the list of the 2026 GRIPS projects. You can apply for one of these projects only.
!!! Please state your preferred project in the motivation letter. !!!
=====================================================
Project 1: Identification and Optimization of chloride storage from green power
Application Partner: MODAL Energy Lab
Safe mass storage of chloride in ionic fluids holds significant potential for temporal energy shifting, enabling optimal use of fluctuating green energy sources. This project focuses on simulation as well as parameter identification and optimal control methods for a chemical reactor system that stores chloride in an ionic fluid. The system integrates fluid flow, species and heat transport, and chemical reactions. Building on a preliminary sensitivity analysis, the goal is to develop and implement simplified models that facilitate parameter identification from experimental or simulated data and to optimize the reactor for both efficiency and throughput. The project requires mathematical techniques such as PDE calculus, finite difference or finite element discretizations, and gradient-based or gradient-free optimization methods. Proficiency in Python and/or MATLAB/Octave is advantageous. By bridging mathematical modeling, numerical simulation, inversion, and optimization, this project offers novel insights into technological aspects of the green energy transition.
Expected outcomes
A prototype code that allows investigating chloride storage mechanisms and options. It shall allow comparing different setups, parameter studies, and finding the optimal choice of selected parameters.
Methodology
-
Modeling of relevant physical processes (reactions, laminar fluid flow, species and heat diffusion and transport, buoyancy) in rotationally symmetric storage reactors
-
Implementation of a simple finite difference forward simulation code in Matlab/Octave/Python/...
-
Computing maximum posterior point estimates for selected parameters from actual and simulated data using gradient-free optimization
-
Optimizing reactor parameters for maximum storage rate
Requirement
-
Interest in numerical simulation and optimization of physical phenomena
-
Programming skills in Matlab, Octave, Python, C++, or similar languages
-
Basic knowledge of numerics for partial differential equations and optimization
==================
Project 2: Task-Specific Foundation Models for Cryo-Electron Tomography
Application Partner: Thermo Fisher Scientific
Cryo-electron tomography (cryo-ET) enables visualization of macromolecular complexes in their native cellular context, but interpretation remains challenging due to high noise levels, the missing wedge artifact, and a lack of ground-truth data. In this project, we take advantage of a previously established foundation model trained by self-supervised representation learning on cryo-ET datasets. This model comprised of Siamese and Feature Pyramid network architectures and is trained to learn a hierarchical representation of tomographic data using a synthetic dataset that models realistic image acquisition parameters. The model achieves promising results on experimental data without fine-tuning and supports key aspects of cryo-ET data analysis, including tomogram denoising, semantic and instance segmentation of subcellular structures across both prokaryotic and eukaryotic biological systems. In this project, we aim to adapt this foundation model toward more task-specific outputs like particle picking and membrane segmentation to help end-user biologist with separate data processing pipelines via a unified model. To this end, we will develop new segmentation and classification heads for the model. We will use a cryo-ET simulator software to generate a diverse dataset including a large number of macromolecular and other cellular structures. Then, we will integrate this synthetic data with unannotated experimental data to fine-tune the model for the specific tasks. The expected output is a robust model tailored for different stages of the cryo-ET workflow. In parallel with these efforts, we attempt to solve the domain gap between synthetic and experimental data using a task-agnostic model and domain adaptation (DA) methods for particle picking downstream task. Here, the model is a 3D Unet that will be trained using synthetic data for particle picking. As there is a domain gap between synthetic and experimental data, this model will perform poorly on experimental data. We will solve this domain shift problem using a DA method with a specific training strategy. Eventually, we will compare the adapted task-agnostic model and adapted foundation model results to gain realistic insights about capabilities and applications of these models in industry.
Requirements
-
Experience with deep neural networks
-
Knowledge and experience with image processing
-
Good programming skills in Python and Pytorch AI framework
==================
Project 3: Mathematical Models of Deformation and Articulation in Model to Image Registration
Application Partner: Stryker Berlin GmbH & Arcus Clinic Pforzheim
This project investigates the foundations of deformable 2D-3D geometric registration for orthopedic analysis. The primary objective is to align three-dimensional bone geometries, either reconstructed from computed tomography data or derived from learned shape priors, with clinically acquired X-ray images, while explicitly modeling physically plausible articulations (joints) and a combination of rigid (osseous) and non-rigid (soft tissue) deformations.
At its core, the project addresses a highly ill-posed inverse problem that couples projective geometry, differential shape analysis, and constrained optimization. Particular emphasis is placed on developing and analyzing mathematical formulations for deformable shape registration in projective spaces, with attention to identifiability, regularization, and physically meaningful constraints. A central challenge is how rigorous geometric and variational principles can be embedded into modern, data-driven shape representations, ensuring interpretability within expressive prior models.
The resulting framework aims to enable quantitative assessment of patient-specific kinematics under varying load-bearing conditions, while also serving as a testbed for theoretical questions under modern geometry in the age of AI.
Requirements
We seek motivated students with a strong interest in the intersection of metric and projective geometry, optimization, and modern geometric modeling, as well as a broad enthusiasm for problem solving at the interface of theory, computation, and emerging shape representations. In particular:
-
Programming skills in Python
-
Familiarity with optimization algorithms
-
Familiarity with projective geometry OR fundamentals of computer vision
-
Familiarity with differential geometry