Join us at the TUM Open House Research Campus Garching!
We’re excited to invite you to our booth at the TUM Open House on October 3rd from 10 AM to 5 PM. Meet relAI students and coordinators at our booth, where you can experiment with building your own neural network or test your intuition against Simpson’s Paradox! You’ll find us in the Magistrale of the Mathematics & Informatics Building, right next to the info point.
At 2:45 PM, relAI Director Stephan Günnemann will deliver a talk on "Wie Künstliche Intelligenz die Wissenschaft verändert" ("How Artificial Intelligence is Changing Science"). This presentation, in German, will take place in the Friedrich L. Bauer Lecture Hall within the Mathematics and Computer Science building.
This Open House is a fantastic opportunity to explore other research institutes, labs, and take part in various hands-on activities throughout the day!
For more details, including how to engage with other activities at the event, visit:
We are thrilled to share the outcome of the recent (03 - 05.07) relAI retreat, which brought together more than 50 fellows and students in Miesbach. Over the course of three days, participants engaged in intriguing discussions, insightful keynotes, and a variety of activities focused on the various aspects of reliability in AI.
Matthias Althoff “Provably Safe Reinforcement Learning through Online Verification” Stefan Feuerriegel “Causal ML for Predicting Treatment Outcomes”
Students got to know each other better during the Speedgeeking and Lightning Talks sessions. The first one consisted of five minutes one to one discussions on questions relevant to relAI students. The Lightning Talks, one minute presentations of each student´s research topic, allowed for a fast though comprenhesive view of the research carried out throughout relAI.
SpeedgeekingLightning talkLightning talk
A part of the retreat followed an unconference format, fostering engaging group discussions on various challenges of reliability of AI. These ranged from the lack of sufficiently large datasets in medicine and healthcare, to explainability of foundation models and the societal impacts of large language models. Some of the discussions focused on improving the relAI program, such as our relAI seminar, relAI blog and relAI safety hackathon.
Group discussionGroup discussionGroup discussionPresentations of a group discussionPresentations of a group discussionPresentations of a group discussion
Towards the end of the retreat, relAI Fellows celebrated the third relAI Fellows assembly, a successful meeting that tackled organizational and research issues of the school.
As relAI integrates into the Universities of Excellence, TUM and LMU, bringing together members who work and study across various parts of Munich, events like this retreat are invaluable. They strengthen the "relAI family," fostering collaboration, scientific exchange and community spirit. We are already looking forward to the next retreat!
Thank you to everyone who participated and contributed to making this event a great success.
/ mc
Save the date for the next Women in AI & Robotics community meetup co-organized by relAI!
On May 14th, our co-director Gitta Kutyniok will talk about “Reliable AI: From Mathematical Foundations to Neuromorphic Computing”. The second presentation of the event will be held by Elke Wolf, Professor from Hochschule München University of Applied Sciences, on “Professor at a University of Applied Sciences – an attractive option for empowered women”.
See you there! Admission is free, but signing up required.
/ mc
Are you interested in frontier AI systems, their astonishing capabilities and risks for humanity? Then join us for a thought-provoking deep dive and exclusive OpenAI Live Q&A on AI safety.
Date: Wednesday, May 8th, 2024 | 19:00 – 20:30
Location: Room B006, Department of Mathematics (Theresienstr. 39) or online
Language: English
Agenda:
19:00 – 19:05: Doors open
19:05 – 19:30: Introduction to AI Safety
19:30 – 20:15: Presentation & Live Q&A with OpenAI researcher Jan H. Kirchner, co-author of weak-to-strong generalization paper
20:15 – 20:30: Closing talk – What can we do?
20:30 – onward: Optional socializing and small group discussions with free drinks and snacks.
Please register on the following webpage and prepare your questions!
/ mc
We are thrilled to share the outcomes of our recent student-driven event organized by Maria Matveev and Julius Hege from the Chair of Mathematical Foundations of Artificial Intelligence (LMU): the first relAI Safety hackathon held last weekend! This dynamic gathering brought together a mix of students and professionals interested in the field of AI Safety.
Over the course of two intense days, participants delved into practical projects aimed at addressing various aspects of AI safety. Their projects ranged from adversarial prompting on a binary question data set, measuring the robustness of the responses, to a website to compare your own emotional intelligence and bias to large language models such as Llama and ChatGPT. The latter project is publicly available, and you can try it out here: mindmatch.streamlit.app
The atmosphere at the hackathon was inspiring, with enthusiastic participants exchanging ideas, insights and experiences on how to enhance the reliability and safety of AI. The event provided a great opportunity for attendees not to only work on innovative projects, but also to engage in thought-provoking discussions surrounding the ethical implications and potential risks associated with AI. Looking forward to more engaging events!
This strategic plan aims to significantly advance AI research, development, and application in Germany. It emphasizes strengthening Germany's position in the global AI landscape. This is done by enhancing research infrastructure and transfer, facilitating AI integration into various sectors such as healthcare, fostering a dialogue on the societal and ethical implications of AI and ensuring AI competencies at all levels in the long term. Notably, the BMBF has committed over 1.6 billion Euros to this AI action plan.
Being explicitly named in this influential plan highlights the dedication of the three Zuse Schools including relAI, to fostering ethical, reliable AI innovations and preparing the next generation of AI leaders.
A huge thank you to our supporters, partners, fellows, and students for making this milestone possible!
/ mc
Day 2 of the 2023 relAI retreat combined inspiring keynote talks by relAI fellows Sandra Hirche (“Learning Controls with Guarantees”) and Johannes Maly (“Implicit regularization of training algorithms and resource efficiency”) with more sessions of lightning talks and two rounds of group discussions, a new activity introduced for the first time during this year's retreat.
Keynote talk by Prof. HircheProf. Maly´s keynote talk
The first group discussion session dealt with research topics suggested by fellows and students. The presentations of the discussions’ outcomes showed the intensity and depth with which these discussions had been conducted. The second round of discussions focused on relAI organizational topics such as the relAI Blog and Alumni and were very productive. They were presented on the last day of the retreat with amazing outcomes, including a newly founded student committee to coordinate the new relAI blog, and a student representative responsible for building a relAI alumni network. The election of new student representatives and IDP discussion sessions rounded off two very successful days.
First group discussionsFirst group discussionsSecond group discussionsIDP discussion sessionLightning talkLightning talk
Although the weather was very much November-like, everyone enjoyed the talks, fruitful discussions and social activities in this great location. The preparations for the next retreat have already started; it will take place in summer 2024 and we’re looking forward to gathering the relAI family again.
/ mc
From November 16th to November 18th, the relAI family gathered at the Yachthotel Chiemsee for its second retreat.
The retreat started on the 16th with a warm opening by our two directors, Stephan Günnemann and Gitta Kutyniok. Afterwards, the relAI fellow Debarghya Ghoshdastidar gave the first keynote talk about “Over-parameterization and Over-fitting: Myths, theory and tools” followed by a very lively discussion around this topic.
Keynote talk by Prof. Ghoshdartidar
Day One´s program concluded with a highlight of this year's retreat, a round of lightning talks, where each of our students introduced themselves and their research focus in one minute. Spread over the two days of the retreat, the one minute presentations together with the social interactions resulted in everyone having a very good idea about what interests all the other students have.
Lightning talksSocial Interactions
/ mc
On the 11th and 12th of October, relAI welcomed the new cohort of relAI Master and Doctoral students. The event included informative sessions about relAI and networking activities. At the Munich Data Science Institute (MDSI), TUM, the relAI directors and coordinators presented the relAI program to the new students. The first relAI cohort of students organised a lively interactive session (photo) to welcome and get to know the new students.
/ mc
On 14 September, the Munich Data Science Institute (MDSI, TUM) celebrated its official opening at the Science Congress Center in Garching. The event included inaugural words by the Bavarian State Minister for Science and the Arts, M. Blume, and theTUM President, Prof. Dr. T. Hofmann. Top-class scientific lectures from MDSI Members, presentations of outstanding MDSI projects, a panel discussion, and a poster session rounded the program.
The Konrad Zuse School of Excellence in Reliable AI (relAI), a joint project between TUM and LMU funded by DAAD, which had been initiated through an MDSI Focus Topic, was one of the projects highlighted at the opening. relAI Coordinator Dr. Andrea Schafferhans introduced relAI to the audience, giving an excellent overview of its educational and research programs as well as its people. relAI was also widely represented at the poster session by relAI Master students and PhD researchers from the TUM University, who discussed their research projects with the numerous guests of the event, among them, MDSI partners and representatives from industry and other research institutions.