Jump to the content
- {{#headlines}}
- {{title}} {{/headlines}}
Contact
Press, Communications and Marketing
Tel.: +49 228 833-144
Fax: +49 228 833-441
presse[at]avh.de
Computer vision, computational linguistics and machine learning
In the world of AI research, Marcus Rohrbach specialised in multimodal learning at an early stage, establishing himself as a world leader with his groundbreaking results. His research, combining image and language, is located at the interface of deep learning and logical reasoning.
Language is a human’s most important tool for communicating what we see. If an intelligent system is to be able to communicate with humans effectively it must be able to relate information in words and sentences to the visual world. This becomes relevant, for example, in generating image and video descriptions. So far, models had achieved promising results in labelling images. Intelligent systems were therefore trained using paired image-sentence datasets and thus could only describe the objects in the context they had encountered them during training. By contrast, Marcus Rohrbach and his team developed a model that can generate descriptions that explain previously unknown objects and their interaction with other objects in an image.
In addition to progress of this kind in the reliability of image recognition models and the development of methods for new and unusual training data, Marcus Rohrbach’s research has also been highly successful in symbol-based reasoning and the reduction of forgetting in continual learning. His models, software and data, which he shares with the worldwide AI community, are widely used in research and industry. His interests include computer vision, computational linguistics and machine learning as well as the question of how models can be made so reliable that humans can trust them.
TU Darmstadt focuses on AI as a core scientific discipline and is a member of both major European AI initiatives, ELLIS and CLAIRE. The strategic focus of AI research at the university is the development of intelligent systems for continuous, safe interaction with humans that accord with the European value system. With his abilities to integrate symbolic AI approaches in deep neural networks, Marcus Rohrbach is supposed to help realise this goal, not just at the university, but nation- and Europe-wide, as a professor for “Multimodal Reliable AI”.
Brief bio
Marcus Rohrbach completed his doctorate at the Max Planck Institute for Informatics in Saarbrücken in 2014 before moving to the United States. He originally worked at UC Berkeley and then, from 2017, continued his research at Facebook AI Research (FAIR), today Fundamental AI Research (FAIR) at Meta AI. His research results are published in several fields of AI including the most important conferences and journals for computer vision, computational linguistics, machine learning and artificial intelligence. Moreover, in 2016 and 2018, he won the most prestigious multimodal learning competition, Visual Question Answering Challenge, as well as the 2018 VizWiz Challenge that concentrates on the challenges facing people with visual impairments.
