Manuel Brack
Machine Learning Group, Computer Science Department, TU Darmstadt. Hochschulstrasse 1, Room S1|03 076, 64289 Darmstadt, Germany
brack (at) cs (dot) tu-darmstadt (dot) de
Deutsches Forschungszentrum für KI, TU Darmstadt.
Landwehrstraße 50A, S4 | 23, 1. OG, 64293 Darmstadt, Germany
manuel (dot) brack (at) dfki (dot) de

Mission. My research focuses on large-scale generative models at the intersection of natural language and computer vision. Consequently, a lot of our recent efforts have been centered around (text-)conditioned image generation, vision-language models, and LLMs. I am particularly interested in investigating the capabilites, limitations and societal impacts of these models, and how to extert more semantic control over the generated content.


Open Thesis/Hiwi. We work with a variety of deep learning models and architectures with our recent efforts being focussed on large-scale multimodal models such as Vision-Language models or generative diffusion models. We have several ideas related on investigating the underlying capabilities learned during large-scale pre-training, as well as related problems and possible mitigation strategies. We are always looking for highly motivated students and offer Hiwi positions both paid and for CP, as well as Master Theses. I do not currently offer supervision for Bachelor theses. Experience in Deep Learning and specifically in Python and PyTorch is required. If you have ideas yourself If you are interested in a Hiwi position of thesis supervision, feel free to reach out to me with your current overview of grades and information on previous projects and experience. Note: thesis proposal.


Occiglot. Together with great colleagues and friends from DFKI, hessian.AI, CommonCrawl and others, I co-founded the Occiglot research collective in early 2024. As an academic, non-profit research collective, we are committed to open science and, thus, open-source LLM development. We actively work on making linguistic diversity, multilingualism, and cultural richness for mulilingual LLMs. As such our current efforts focus on large-scale data collection & curation, model training, and evaluation of non-english LLMs. At the same time the Occiglot community is providing a hub for researchers and practitioners to exchange ideas, collaborate, and work on shared projects. Our models and datasets can be found on Huggingface with more information available in our blog.


Timeline.
2024: Research Scientist Intern at Adobe Applied Research, San Jose USA.
2024 - now: Co-Founder of the Occiglot research collective
2023: Research Scientist Intern at Adobe Research, Seattle USA.
2023 - now: Research Scientist at the German Research Center for Artificial Intelligence (DFKI),
Research Department for Foundations of Systems AI (SAINT), Darmstadt Germany.
2022 - now: Ph.D. Candidate at the Machine Learning Lab, CS Department, TU Darmstadt, Germany.
2020 - 2022: M.Sc. (with honors) in computer science from TU Darmstadt, Germany.
2016 - 2020: B.Sc. in computer science from TU Darmstadt, Germany.



Publications


Publications can be found at Google Scholar, DBLP, SemanticScholar