Jeremias Bohn

M.Sc.

Research Interests

In recent years, natural language processing took quite a leap forward and the latest achievements, such as ChatGPT, or NLP-based text-to-image generation such as DALL·E and Midjourney, found massive adoption by the broad public and already play a part in modern pop culture. However, the required large language models come at a high cost: The amount of energy consumed as well as hardware required by training and inference of those is tremendous: According to a Bloomberg article from March 2023, Google alone required 2.3 TWh for its AI department in 2021, which corresponds to approximately 1,200 average one-person households in Germany. Further, that unsatiable hunger for hardware and energy of those recent developments leads to higher barriers for contributing to this fields, shifting major breakthroughs from publicly funded researchers to profit-oriented big corporations.

To mitigate this issue, my research revolves around green and recyclable AI, with the following topics being of particular interest:

• distillation of (large) language models
• efficient parameter updating
• 'up-/recycling' of non-SOTA architectures and pre-trained models
• effective training with sparse data

Supervision of Theses

If you are looking for a thesis or guided research topic and are motivated to work on a project related to my research interests, don't hesitate to contact me via email! Please attach your CV, Transcript of Records, and a short motivational statement (< 400 words). Before sending your application, please note the following:

• I'm less interested in your grades, but rather want to know what courses you took, what projects you worked on, how motivated you are to do research on your own and your experience in NLP and machine learning research (if any).
• As I only have very limited capacities, please understand that I cannot offer a thesis to every interested student. However, I will try to answer your enquiries nevertheless, i.e. if you haven't heard from me, I am still deciding whether I can provide you a topic or not.
• In general, I will not cooperate with companies you (want to) work at as this goes against my personal ethical codex of research. If you are only looking for a TUM advisor for an external (paid) thesis, please refrain from contacting me.

Teaching

• WiSe 23/24: Teaching Assistant ("Übungsleitung") for Discrete Structures (INHN0004)
• SoSe 23: Teaching Assistant ("Übungsleitung") for Computational Mathematics I: Linear Algebra (INHN0009)
• WiSe 22/23: Teaching Assistant ("Übungsleitung") for Discrete Structures (INHN0004)
• SoSe 22: Teaching Assistant ("Übungsleitung") for Computational Mathematics I: Linear Algebra (INHN0009)
• WiSe 21/22: Teaching Assistant ("Übungsleitung") for Discrete Structures (INHN0004)
• SoSe 21: Tutor for Linear Algebra for Informatics (MA0901)
• WiSe 19/20: Tutor for Discrete Structures (IN0015), Tutor for Mathematical Preparatory Course
• SoSe 19: Tutor for Linear Algebra for Informatics (MA0901)
• WiSe 18/19: Tutor for Discrete Structures (IN0015), Tutor for Mathematical Preparatory Course
• SoSe 18: Tutor for Linear Algebra for Informatics (MA0901)