Master Thesis: AI-Powered Simulations for Real-World Medical Applications

18.11.2024, Abschlussarbeiten, Bachelor- und Masterarbeiten

Master Thesis: AI-Powered Simulations for Real-World Medical Applications

Scientific simulations play a crucial role in solving complex real-world problems, from modeling natural phenomena to designing advanced materials. However, traditional numerical simulation methods are often computationally expensive, limiting their scalability and real-time applicability.

Recent advancements in artificial intelligence, particularly deep learning, offer a transformative opportunity to enhance the speed and efficiency of these simulations. By leveraging AI, we aim to approximate intricate computational processes while maintaining high accuracy, enabling faster insights and broader applications. Initial research shows promising results, where neural networks predict complex system behaviors using much less computational power compared to traditional methods.

Your role will be to explore cutting-edge neural architectures and innovative training strategies to develop AI-driven solvers [1, 2] for real-world applications. While our main interest lies in reaction-diffusion equations for brain tumor modeling [3], this project is highly flexible, and the exact focus is relatively open and will be determined collaboratively, also taking your interest into account.

Your qualifications:
• Motivated by tackling real-world challenges with impactful solutions.
• Currently enrolled in computer science, math, physics, or a related discipline.
• Strong programming skills in Python, preferably with frameworks like PyTorch or TensorFlow.
• Prior experience with modern deep learning or scientific computing is a plus.

What we offer:
• A chance to pioneer AI-driven solutions in scientific discovery.
• Close mentorship and access to state-of-the-art hardware resources.
• Collaboration with a team of experts in AI, computational science and beyond.

How to apply:
Send an email to Jonas Weidner j.weidner@tum.de with your CV and transcript.

References:
[1] Alkin, Benedikt, et al. "Universal Physics Transformers: A Framework For Efficiently Scaling Neural Operators." The Thirty-eighth Annual Conference on Neural Information Processing Systems.
[2] ICML 2024 Tutorial"Machine Learning on Function spaces #NeuralOperators" https://www.youtube.com/watch?v=_j7bceE9AyA&t=2517s]
[3] Weidner, Jonas, et al. "Rapid Personalization of PDE-Based Tumor Growth using a Differentiable Forward Model." Medical Imaging with Deep Learning. 2024.

Kontakt: j.weidner@tum.de