PhD Position in Mathematical Foundations of Deep Learning

Explore the theoretical underpinnings of deep learning stability and generalization through random matrix theory and probability.

Best for:Best for mathematicians or computer scientists with a master’s degree in applied mathematics, probability theory, or related fields, passionate about theoretical contributions to deep learning.

TU Darmstadt · Darmstadt, Germany · R1 · First-stage researcher

More professors at TU Darmstadt

random matrix theoryprobability theorydeep learning theoryneural network generalizationstability in large-scale networksfeature learning in machine learning

Deadline

Oct 22, 2026

8 days

Funding

—

Start date

flexible (upon appointment)

Full-time

Location

Darmstadt

Germany

Overview

This PhD position at Technische Universität Darmstadt (TU Darmstadt) focuses on advancing the mathematical foundations of deep learning. The successful candidate will contribute to the Emmy Noether project 'RMT4DL: Random Matrix Theory for Deep Learning,' investigating theoretical approaches to understand stability, generalization, and feature learning in large-scale neural networks. The work will be conducted within the Signal Theory and Statistical Learning group, combining expertise in probability theory and random matrix theory. Applicants should have a strong background in mathematics or computer science and excellent English proficiency. The position offers a full-time, temporary contract with a focus on cutting-edge research at the intersection of applied mathematics and machine learning.

mathematical foundations of deep learningrandom matrix theory applicationsprobability theory in machine learningneural network stabilityEmmy Noether projecttheoretical computer scienceapplied mathematicsmachine learning theory

What you'll do

Develop new theoretical approaches to analyze stability and generalization in large-scale neural networks using random matrix theory and probability theory.
Conduct independent research within the Emmy Noether project 'RMT4DL: Random Matrix Theory for Deep Learning,' contributing to the broader understanding of feature learning mechanisms.
Collaborate with the Signal Theory and Statistical Learning group to integrate theoretical findings with empirical machine learning studies.
Present research findings at academic conferences and publish results in high-impact journals.
Participate in group meetings, seminars, and workshops to foster interdisciplinary discussions and knowledge exchange.

Requirements

Degree required: Master’s degree in Mathematics, Applied Mathematics, Computer Science, or a related quantitative field.

Degree offered: PhD

Language: English; excellent proficiency (e.g., IELTS 7.0+ or equivalent)

Instruction: English

Required skills

Strong background in probability theory and random matrix theory
Familiarity with theoretical aspects of machine learning and neural networks
Excellent mathematical modeling and analytical skills
Proven ability to conduct independent research
Proficiency in academic writing and presentation

Nice to have

Experience with computational tools for mathematical modelingKnowledge of deep learning frameworks (e.g., PyTorch, TensorFlow) for empirical validationBackground in statistical signal processing

Eligibility

No specific nationality restrictions, but eligibility for German public service may apply depending on contract terms
No age restrictions mentioned

Funding & terms

Type

Unknown

Amount

—

Duration

—

Programme

—

How to apply

Deadline: 2026-10-22 (Europe/Berlin)

Applicants must submit their application through the TU Darmstadt career portal. Required documents include a CV, a letter of motivation, academic transcripts, and a research proposal outlining the intended approach to the project. Two letters of recommendation are also required. Further details and the application link are available on the TU Darmstadt career website.

Documents required