|Leader(s):||Gunnar Rätsch, Oliver Stegle , Mihaela van der Schaar|
|Members:||Matthew Blaschko , Christoph Bock , Karsten Borgwardt , Sepp Hochreiter , Lena Maier-Hein , Magnus Rattray , Guido Sanguinetti , Julia Schnabel , Fabian Theis, Jean-Philippe Vert|
- Connect and promote the ELLIS vision within the broad areas of human health
- Demonstrate the impact of AI/ML on biomedicine and health
- Feed back key challenges of health applications into AI/ML methods development
- Foster training and education of the next generation of interdisciplinary scientists at the interface of health and AI/ML
ELLIS Robot Learning: Closing the Reality Gap!
|Leader(s):||Tamim Asfour , Aude Billard , Jan Peters|
|Members:||Oliver Brock, Dario Floreano , Danica Kragic , Manuel Lopes , Gerhard Neumann, Paul Newman , Justus Piater , Ingmar Posner , Davide Scaramuzza , Jürgen Schmidhuber, Carme Torras , Marc Toussaint, Ales Ude , Patrick van der Smagt|
This program focusses on central questions for closing the real-world gap for intelligent systems:
How should the robot move? How to act? How to interact? How can sensorimotor behavior be improved by machine learning approaches?
Geometric Deep Learning
|Leader(s):||Michael Bronstein , Taco Cohen , Max Welling|
|Members:||Pascal Frossard , Marco Gori , Pietro Lio , Frank Nielsen , Frank Noé , Jakob Verbeek , Stefanos Zafeiriou|
- Machine learning on non-Euclidean domains
- 4G: geometric, graph, group, gauge convolutions
- Applications: computer vision, graphics, social networks, chemistry, biology, physics, medicine
Human-centric Machine Learning
|Leader(s):||Plamen Angelov , Nuria Oliver , Adrian Weller|
|Members:||Nozha Boujemaa, Carlos Castillo , Ciro Cattuto , Lilian Edwards , Sergio Escalera , Manuel Gomez Rodriguez , Krishna Gummadi , Bruno Lepri , Chris Russell , Christian Theobalt , Karen Yeung|
- help to ensure widespread benefits to, and acceptance from, the public by guaranteeing:
- transparency, clear accountability, interpretability and fairness of the algorithmic decisions
- amenable to legal and technical certification, accountability and verifiability.
Interactive Learning and Interventional Representations
|Leader(s):||Nicolò Cesa-Bianchi , Andreas Krause , Bernhard Schölkopf|
|Members:||Barbara Caputo , Volkan Cevher , Thore Graepel, Nicolas Heess, Stefanie Jegelka , Christoph Lampert , Timothy Lillicrap , Shie Mannor , Yishay Mansour , Massimiliano Pontil , Doina Precup, Carl Rasmussen , John Shawe-Taylor , Csaba Szepesvari , Chris Watkins , Shimon Whiteson|
We rethink the principles of interactive models of learning, exploring the role of causal modelling in bridging the gap between observational and interventional learning. The ultimate goal is to understand the organizing principles underlying robust intelligent behaviour, and to enable reliable learning-based decision systems for high-stakes real-world applications.
- Principles of learning-in-the-loop systems
- Online and reinforcement learning
- Causal inference
- Interacting learning systems (multi-agent learning, games, networks)
Machine Learning and Computer Vision
|Leader(s):||Bernt Schiele, Cordelia Schmid , Yair Weiss|
|Members:||Zeynep Akata , Michael Black , Thomas Brox, Daniel Cremers, Rita Cucchiara , Andrew Fitzgibbon , Andreas Geiger , Michal Irani , Ivan Laptev , Jiri Matas , Tomas Pajdla , Jean Ponce , Stefan Roth , Josef Sivic , Tinne Tuytelaars, Luc Van Gool , Andrea Vedaldi , Andrew Zisserman|
Computer Vision has been revolutionalised by Machine Learning. Our goal is to connect classical Vision algorithms and modern machine learning more explicitly.
- Mid-level vision and image reconstruction
- 3D Geometry from multiple views
- Object and activity recognition
Machine Learning for Earth and Climate Sciences
|Leader(s):||Gustau Camps-Valls , Markus Reichstein|
|Members:||Nuno Carvalhais , Joachim Denzler , Veronika Eyring , Kristian Kersting , Miguel Mahecha , Jonas Peters , Stephan Rasp , Jakob Runge , Sancho Salcedo , Konrad Schindler , Dino Sejdinovic , Bjorn Stevens , Devis Tuia , Xiangxiang Zhu|
Goal: Model and understand the Earth system with Machine Learning and Process Understanding
- Spatio-temporal anomaly and extreme events detection, anticipation and attribution
- Data-driven dynamic modelling and forecasting
- Hybrid modeling: linking physics and machine learning models
- Causal inference, Learning and explaining feature representations
- Earth and Climate model emulation, generative modelling and data-model fusion
- Benchmark synthetic and real datasets
|Leader(s):||Matthias Bethge , Y-Lan Boureau , Peter Dayan|
|Members:||Tim Behrens , Matt Botvinick, Nick Chater , Emmanuel Dupoux , Sharon Goldwater , Raia Hadsell , Anne Hsu , Bradley Love , Mackenzie Mathis , Josh Tenenbaum , Oriol Vinyals , Jane Wang , Li Zhaoping|
The standard paradigm of machine learning is task-centric.
Natural intelligence is agent-centric: a single brain shaped through evolution learns to perform all tasks.
- Lifelong learning
- Deep semantics and cross-domain learning
- Shaping inductive bias via neural network structure
- Adaptive resource deployment
- Social reasoning
|Leader(s):||Bert Kappen , Riccardo Zecchina|
|Members:||Carlo Baldassi , Giulio Biroli , Guiseppe Carleo , Gabor Csanyi , Vedran Dunjko , Jens Eisert , Florent Krzakala, Florian Marquardt , Miguel Angel Martin-Delgado , Remi Monasson , Matthias Rupp , Giuseppe Santoro , Lenka Zdeborova|
The aim of the Ellis program Quantum and Physics based machine learning (QPhML) is to use concepts from quantum physics and statistical physics to develop novel machine learning algorithms with the ultimate aim to realize novel future, possibly energy efficient, hardware implementations. Objectives:
- Exploit quantum effects in machine learning
- Accelerate and improve energy efficiency of machine learning algorithms through dedicated physical implementations
- Use machine learning methods to advance understanding of quantum information processing
Program website: https://www.snn.ru.nl/v2/lan/en/ellis.content
Robust Machine Learning
|Leader(s):||Chris Holmes , Samuel Kaski , Yee Whye Teh|
|Members:||Cédric Archambeau , Silvia Chiappa , Yarin Gal , Zoubin Ghahramani , Amir Globerson , Peter Grünwald , Frank Hutter , Pushmeet Kohli , Sebastian Nowozin , Antti Oulasvirta , Richard Turner , Isabel Valera , Aki Vehtari , Chris Williams|
- Principles and methods for Robust ML
- Quantification and verification of Robust ML
- Applications in health, environmental sciences, design, autonomous vehicles, industrial control.
Theory, Algorithms and Computations of Modern Learning Systems
|Leader(s):||Francis Bach , Philipp Hennig , Lorenzo Rosasco|
|Members:||Mario Figueiredo , Asja Fischer , Arthur Gretton , Matthias Hein , Martin Jaggi , Julien Mairal , Michael Osborne , Simo Särkkä , Thomas Schön , Ingo Steinwart , Naftali Tishby , Ulrike von Luxburg|
- Many contemporary ML algorithms are still comparably badly understood conceptually. As a result, they require manual tuning, can be inflexible and behave erratically.
- The program connects experts with diverse backgrounds to advance the algorithmic foundations of ML. It will support the development of efficient and reliable learning systems with theoretical guarantees.