From July 28 to July 31, 2026, relAI hosted the second relAI International Summer School at LMU. This event focused on promoting educational exchange aimed at developing reliable AI and exploring current trends in the field.
Last week, our industry partner, QuantCo, hosted a networking event with relAI at their new Munich office.
The relAI students had a wonderful evening filled with engaging conversations and delicious food. They had the opportunity to meet colleagues from QuantCo's AI, and engineering teams and learned about how the company develops scalable systems to transform messy, real-world data into impactful decisions.
We are pleased to share relAI’s contribution to the European Embodied Robotics Week, whichtook place last week in Munich. The event was organizedby RoboTUM together with START Munich and the ESRA (European Student Robotics Association) network.
The Robotics Festival convened a diverse range of stakeholders from the European robotics and physical artificial intelligence ecosystem. Specialists, students, and robotics enthusiasts participated in city-wide events, including open houses at robotics laboratories, makerspaces, and studios throughout Munich, as well as a hackathon.
The event ended with a summit featuring talks and panels from leaders in embodied intelligence. relAI was represented in the panel discussion “Dexterity and Manipulation for robotics” by relAI Fellow Prof. Khadiv.
/ mc
On June 8, 2026, TUM hosted the International Symposium 'Future Learning: Global Perspectives, European Pathways' as part of the new relAI research area Learning & Instruction.
Are you interested in contributing research for underserved communities? Don't miss the EEAMO Conference 2026, the 6th ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization.
This event will showcase work across the research-to-practice pipeline, aiming to ensure that algorithmic systems serve a broadly beneficial role in society by advancing equity and expanding access to opportunities for underserved communities.
On January 20, 2026, researchers from relAI, MDSI and our industry partner SAP gathered for a pitchtalk session at the SAP Labs Munich Campus in Garching. In an interactive setup, participants from the three organizations exchanged research insights and explored ideas for future collaborations.
The afternoon opened with short welcome words from representatives of SAP (Dr. Tobias Müller), MDSI (Sylvia Kortüm) and relAI (Dr. Mónica Campillos). The following pitchtalks covered a wide range of topics relevant to AI and Data Science, spanning methodological approaches to practical research resources.
Pitchtalk Session
The talks introduced methodologies and applications across various research fields, from uncertainty modeling to personalized medicine and research data infrastructure.
Max Beier, a relAI PhD student, discussed the advantage of using linear-operator methods in dynamical systems. By focusing on linear operators, his approach supports optimal control, scalable optimization algorithms, and reliable forecasting across time scales ranging from milliseconds to days. He showed that this methodology allows efficient learning of system behavior, modeling of trajectory distributions, and generation of coherent sequences, addressing current limitations in decision‑making models for complex dynamical environments.
Parastoo Pashmchi, an industrial PhD student at SAP, introduced an efficient algorithm to handle missing data, a common challenge in AI projects. She presented an ML‑based imputation method that preserves the original data distribution by sampling from the conditional distribution of nearest neighbors, overcoming the limitations of common techniques like kNNImputer. By enabling uncertainty quantification and multiple imputations, her approach improves the reliability of predictive models such as SAP’s green energy forecasting use case, where missing solar production data can significantly undermine model accuracy.
Mario Picciani, an MDSI PhD student, highlighted recent developments and applications of ProteomicsDB, a proteomics resource established in 2012 through a collaboration between MDSI Core Member Prof. Bernhard Küster and SAP. ProteomicsDB is a powerful multi‑omics, multi‑organism platform that enables real‑time exploration of proteomic, transcriptomic, and drug‑interaction data across the tree of life, supporting research from basic biology to large‑scale initiatives. With expanding capabilities for analyzing drug mechanisms, predicting cell responses, and supporting precision oncology, the SAP HANA–powered resource could become a central tool for biomarker discovery, systems biology, and personalized medicine.
Sebastian Gallenmüller, an MDSI PhD student, presented SLICES-DE, a digital research infrastructure for computing and communication, embedded within a European collaborative network. As a national digital research infrastructure for ICT, SLICES‑DE offers remote‑accessible testbeds, reproducible workflows, and long‑term data management to support research in areas such as 6G, AI, cybersecurity, and cloud‑edge systems. Built as a community‑driven, flexible, and scalable platform, it enables shared experiments, training, and industry collaboration, providing both academia and companies with a versatile environment that can even be booked for individual lectures or large‑scale projects.
Networking over pretzels and lemonade
An informal networking session after the pitchtalks gave speakers and participants from relAI, MDSI, and SAP the opportunity to connect, exchange impressions, discuss research interests, and explore potential collaborations.
We thank our industry partner SAP for hosting this event and sharing insights into ongoing research projects – we look forward to future editions!
We are pleased to announce the first Munich AI Lecture of 2026, presented by Prof. Dr. Sebastian Pokutta on the growing role of Artificial Intelligence in scientific discovery. The lectureis organized by the Bavarian AI Chair for Mathematical Foundations of Artificial Intelligence at LMU, led by relAI Director Gitta Kutyniok.
Details:
Title: How Machines Explore, Conjecture, and Discover Mathematics
📍 Geschwister-Scholl-Platz 01, Raum D209, 80539 München, Deutschland
The lecture focuses on the role of AI as a partner in mathematical research. Prof. Pokuttawill present approaches from the AI4Math initiative that combine optimization, machine learning, and mathematical structure to open up highly complex search spaces. Using the Hadwiger–Nelson problem as an example, he will explain how neural networks can be used to convert mixed discrete-continuous problems into differentiable optimization problems and explore new solution spaces.
About the Speaker
Prof. Dr. Sebastian Pokutta is Vice President of the Zuse Institute Berlin (ZIB) and a Professor of Mathematics at the TU Berlin, with a research focus on Artificial Intelligence and Optimization. He leads, among other initiatives, the Excellence Cluster MATH+ and the Research Campus MODAL and has previously worked in academia and industry, including at MIT, IBM ILOG, and Georgia Tech. His work has been recognized with numerous awards, such as the Gödel Prize, the STOC Test of Time Award, and the NSF CAREER Award.
Join us for a lively meetup featuring leading researchers from MDSI, relAI, and experts from relAI Industry Partner SAP, where we will connect, discuss, and explore new ideas.
🗓️ January 20, 2026, 4:00 - 5:30 pm (doors open at 3:30 pm, open networking after the talks)
📍 SAP Labs Munich Campus (MUE03), Friedrich-Ludwig-Bauer-Straße 5, 85748 Garching bei München, Auditorium (AE.76)
Max Beier (relAI): Learning, Optimization, and Uncertainty in Dynamical Systems with Linear Operators Dynamical systems are ubiquitous in data science and robotics, motivating a broad range of methods for learning models from data, optimizing their behavior, and quantifying uncertainty in their predictions. This talk will showcase how linear-operator methods provide a unified way to study learning, optimization, and uncertainty in dynamical systems. Operator learning treats both inputs and outputs as functions and learns mappings between function spaces, which reframes dynamical modeling as the identification of operators from data. Restricting attention to linear operators enables compact representations, output-independent representations, and efficient computation through the use of linear algebra. We will showcase how this framework can be successfully employed to: learn a representation of a dynamical system for optimal control, model distributions over trajectories efficiently, and generate coherent sequences.
Sebastian Gallenmüller (MDSI): SLICES-DE: Digital Research Infrastructure for Computing and Communication SLICES-DE (Scientific Large-scale Infrastructure for Computing/Communication Experimental Studies - Germany) is a digital research infrastructure and Germany’s national node of the European SLICES-RI initiative. It provides a remote-accessible experimental platform for researchers in information and communication technologies (ICT) with structured workflows to support reproducible experiments and long-term data archiving. The infrastructure is designed to be flexible and scalable to a wide range of research domains, including 6G networks, AI, cybersecurity, and cloud-to-edge systems. SLICES-DE aims to provide a research infrastructure for ICT research in Germany, with engagements possible for academic and industry partners.
Parastoo Pashmchi (SAP): Handle your Missing Values Easily: An ML-powered Solution for Filling Data Gaps Every machine learning and artificial intelligence model relies heavily on data, and managing this data, especially when it comes to dealing with missing values, presents a significant challenge for data scientists. Missing data can lead to a loss of vital information needed for training models and can disrupt the patterns or relationships that prediction algorithms rely on. For example, incomplete financial datasets can yield flawed risk assessments, while missing information in clinical trials or engineering sensor data can compromise the validity of study outcomes and fault detection, respectively. Similarly, missing weather data can lead to inaccurate green energy generation predictions, affecting grid management and planning. Our proposed method provides a fast and easy-to-implement algorithm for imputing missing values, emphasising the preservation of the original data distribution and pattern based on the available data. This algorithm overcomes some of the limitations of conventional techniques and provides a theoretical guarantee for effectively recovering the distribution of the missing values. Key features of this algorithm include the quantification of uncertainty for each imputed value and the capability for multiple imputations to consider various potential scenarios. These features enhance the reliability and stability of our machine learning model, offering more reliable predictions. In this session, we will highlight the algorithm's main features and demonstrate its application in predictive models.
Mario Picciani (MDSI): ProteomicsDB: A Multi-Omics and Multi-Organism Resource for Life Science Research ProteomicsDB is an evolving, publicly accessible multi-omics database originally developed to enable the interactive exploration of large quantitative mass spectrometry–based proteomics datasets, which led to developing the first draft of the human proteome. Initially, the HANA-powered platform supported the real-time comparison of protein abundance across human tissues, cell lines, and body fluids, but was subsequently expanded to integrate other omics layers such as transcriptomics, drug-target interactions, protein–protein interactions, and cell viability data, transforming it into a versatile resource for multi-omics life science research. Over further successive updates, ProteomicsDB was expanded to support additional organisms covering nearly all branches of the tree of life facilitating research towards the concept of “One Health” - integrating and broadening our understanding of the interactions between humans, animals, plants, microorganisms, soil, and the environment to ensure a sustainable and resilient future of healthy living. The latest update strengthened its role to facilitate the comprehensive analyses of drug mechanisms, cell sensitivity profiles, and biomolecular signatures to support precision medicine applications. ProteomicsDB’s rich data content makes it an ideal resource for basic to translational research, with applications in systems biology, and drug discovery, and its versatile and extendible architecture makes it an ideal platform for large-scale research initiatives, illustrated by its central role in projects such as ‘The Proteomes That Feed The World’ and ‘Reducing Non-Human Primates in Non-Clinical Safety Assessment’.
