ELLIS Unit Vienna: Committed to Interdisciplinarity and Academic Freedom

The ELLIS Unit Vienna, based at the Institute of Science and Technology Austria (ISTA), stands out for its interdisciplinary approach and broad scope. As part of the ISTA ecosystem, it is dedicated to excellence in advancing both the foundations and scientific applications of machine learning (ML).

Mission & Vision 

The ELLIS Unit Vienna, based at the Institute of Science and Technology Austria (ISTA), is committed to advancing machine learning (ML) research and its interdisciplinary applications across various scientific domains. Their mission emphasizes academic freedom, fostering an environment where curiosity-driven research thrives, leading to scientific excellence.

With ELLIS, we seek to emphasize academic freedom in AI advancement. We are persuaded that scientific excellence flourishes in curiosity-driven research environments. - Christoph Lampert, Director ELLIS Unit Vienna

Key Numbers and Milestones

Early adopter: The ELLIS Unit Vienna was among the very first ELLIS Units to join the ELLIS network in 2019, marking a significant step in advancing AI and machine learning research in Europe. Its establishment was initiated by Unit Director Christoph Lampert, laying the foundation for a hub dedicated to excellence, interdisciplinary collaboration, and groundbreaking innovation in the field.

Growth in Members: The ELLIS Unit Vienna has grown to include eight core faculty members, including Alex Bronstein, Christoph Lampert, Dan Alistarh, Francesco Locatello, Marco Mondelli, Matthew Robinson, Monika Henzinger, and Vladimir Kolmogorov.

Strong research funding: The unit's core faculty has independently secured multiple ERC grants, including six active since 2019, along with industry funding.

Event hosting: In July 2024, Vienna hosted the International Conference on Machine Learning (ICML 2024), a premier event in the AI and ML community. Seizing this opportunity, the ELLIS Unit at ISTA organized the "Future of Machine Learning Symposium" on July 19, 2024, just two days prior to ICML. This symposium convened leading academics and industry experts at ISTA to discuss recent advancements and emerging trends in machine learning, providing a platform for knowledge exchange in a more intimate setting compared to the larger conference environment.

Research Areas 

Alex Bronstein - Computational Imaging and Machine Learning Applications in Structural Biology

The Bronstein group focuses on developing computational data-driven methods to address complex challenges in natural and life sciences and engineering. Their research encompasses computational imaging, where they integrate optics design, sensor technology, and advanced algorithms to optimize imaging systems for specific tasks. Additionally, the group applies and advances machine learning techniques for structural biology, particularly in modeling, analyzing, and designing protein structures, dynamics, and functions.

Christoph Lampert - Machine Learning and Computer Vision

The Lampert group focuses on developing trustworthy machine learning algorithms with an emphasis on robustness, fairness, and privacy. Their goal is to ensure that machine learning systems behave predictably and safely, even under challenging conditions or when deployed in sensitive real-world scenarios. Additionally, they explore transfer learning, including continual, meta, and lifelong learning, to enhance the adaptability and reliability of machine learning models.

Dan Alistarh - Deep Algorithms and Systems Lab

AI has seen major advances, but its progress is challenged by the exponentially growing computational costs of training and deployment, limiting accessibility and innovation. The Alistarh group advances AI democratization by developing efficient training and inference algorithms, focusing on compressed representations and optimized system implementations.

Francesco Locatello - Causal Learning and Artificial Intelligence

The Locatello group focuses on causal learning, enabling AI models to discover causal structures in data. Their group pioneered causal representation learning and made key contributions at the intersection between causality and machine learning. Ongoing research includes representation learning for causal inference and causal discovery, generalization in deep learning, and AI-driven discovery of differential equations to describe physical phenomena. They apply their research to a number of scientific disciplines, including virtual cells, climate, and experimental sciences.

Marco Mondelli - Data Science, Machine Learning, and Information Theory

The Mondelli group tackles complex inference problems across fields like wireless communications and machine learning, developing mathematically grounded solutions inspired by information theory. They aim to determine the minimal data needed for a task, design efficient algorithms, and explore trade-offs between factors like data size, complexity, and performance.

Matthew Robinson - Modeling Large-Scale Medical Record Data

The Robinson group develops statistical models and computational tools to analyze large-scale human medical records, aiming to understand how genetics and lifestyle shape disease risk. Their research focuses on why symptoms appear at different ages, vary in severity, and how diseases are interconnected. They also study key life stages, including pregnancy, growth, and aging, to improve disease prediction and treatment strategies.

Monika Henzinger - Privacy-Preserving Algorithms

The Henzinger group investigates algorithms focused on privacy-preserving techniques for input data. Their research includes designing methods for processing and analyzing large-scale dynamic graphs while ensuring the privacy of sensitive information. This involves developing secure data structures and algorithms that allow for data analysis and decision-making without exposing individual user data, thereby maintaining privacy in applications like social networks, healthcare, and financial systems. Their work aims to strike a balance between computational efficiency and the protection of private information in real-time processing.

Vladimir Kolmogorov - Discrete Optimization

The Kolmogorov group focuses on three main areas: efficient algorithms for inference in graphical models and combinatorial optimization, theoretical investigations into the complexity of discrete optimization, and applications in computer vision. Notable contributions include the "Boykov-Kolmogorov" maximum flow algorithm, the "TRW-S" algorithm for MAP inference, and the "Blossom V" algorithm for minimum cost perfect matching. Their work also applies discrete optimization techniques to image segmentation and stereo reconstruction.

Ongoing Collaborations

The ELLIS Unit Vienna is deeply integrated into ISTA's innovation ecosystem, collaborating closely with xista, ISTA's hub for science-based innovation and tech transfer. This collaboration was highlighted at the bigX conference, held on September 19, 2024, which focused on “The Future of AI”. The event attracted visionary researchers from institutions like MIT, Oxford, and UMass, as well as industry leaders such as Novartis, Siemens Energy, and AOP Health.

Being part of this ecosystem facilitates connections between the ELLIS Unit Vienna and numerous ML companies and startups in and around Vienna. Members also have access to advanced computational resources, including a state-of-the-art NVIDIA GPU cluster for generative AI and machine learning. This infrastructure supports research across various fields, primarily benefiting the Unit's faculty while also enabling applications in life sciences, physics, and computer science.

Additionally, five ISTA professors, including four Members of the ELLIS Unit Vienna, are leading the “Bilateral AI” Cluster of Excellence. This initiative aims to integrate sub-symbolic and symbolic AI approaches. The Cluster, involving six Austrian institutes and universities, has secured €33 million in funding over five years, with 60% provided by the Austrian Science Fund (FWF). 

Supporting and Fostering Young Talent

ELLIS PhD student Riccardo Caddei’s project, Smoke and Mirrors in Causal Downstream Tasks (arXiv preprint), examines how biases can arise in downstream tasks despite a well-understood causal model. The research, led by Locatello at ISTA, involved collaboration with Cordelia Schmid (Inria, France) and social immunity expert Sylvia Cremer (ISTA). This project exemplifies the strength of the ELLIS PhD program, fostering high-impact research through European collaboration and interdisciplinary exchange.

The prestige aspect of ELLIS gives us a competitive advantage in hiring graduate students in AI and ML. - Francesco Locatello, Deputy Director, ELLIS Unit Vienna

ELLIS Engagement

Meet & Greet: The ELLIS Unit Vienna organises activities such as on-campus networking events and the ELLIS Unit Lunch to foster collaboration among AI and ML researchers.

Regular ELLIS Talks by Renowned International Researchers: The ELLIS Unit Vienna hosts a series of talks featuring distinguished researchers from around the globe. These sessions provide insights into cutting-edge advancements in artificial intelligence and machine learning, fostering knowledge exchange and collaboration within the academic community.

Presentations by ELLIS PhD Students at Local and International Meetings: PhD students affiliated with the ELLIS Unit actively engage in presenting their research at various platforms.