ELLIS PhD Award

ELLIS has established the ELLIS PhD Award to recognize and encourage outstanding research achievements during the dissertation phase in artificial intelligence and machine learning related fields (including computer vision and robotics). The ELLIS PhD Award is sponsored by the Kühborth Stiftung GmbH. Each awardee will receive Euro 2,500 prize money. Typically, there will be two awards per calendar year, one given to a female researcher and one to a male researcher.

Learn about the recipients of the ELLIS PhD Award in 2024.

How to Nominate an Outstanding Researcher

The nomination must include:

A short summary (1-3 pages) of the main research achievements of the dissertation
2-3 of the most important publications resulting from the dissertation
A scan of the dissertation certificate (which includes the date of the defense)
Final version of the dissertation as pdf (or link to pdf)
The candidate's CV
A short laudation (40-80 words)

Please use this template (pdf) (latex) to nominate a candidate and send the nomination (preferably as a single pdf file) to phd-award@ellis.eu. The awards will be given during an ELLIS event in 2025. If you should have any questions, please send an email to phd-award@ellis.eu.

Please note that self-nomination by the author of the dissertation is not allowed.

Deadline

The deadline for nominating a dissertation for the year 2024 is:

April 15, 2025 (End of Day, Anywhere on Earth)

Eligibility

Any European dissertation in the area of artificial intelligence and machine learning related fields (including computer vision and robotics) that has been defended in 2024 can be nominated. A dissertation is considered European if the dissertation has been performed primarily at a European research institution.

Selection Criteria

Dissertations will be reviewed for technical depth and significance of the research contribution and potential impact on theory and practice.

Kühborth Stiftung GmbH

Kühborth Stiftung GmbH is a foundation established by Dr. Dipl.-Ing. Wolfgang Kühborth and his wife Helga Kühborth to promote research and teaching in the fields of natural, technical and economic sciences. It was for this purpose that the founders transferred their shares of Klein Pumpen GmbH (today: Johannes und Jacob Klein GmbH) to the Kühborth Stiftung in 1994. Johannes und Jacob Klein GmbH holds more than 80 percent of the ordinary shares of KSB SE & Co. KGaA, one of the world's leading manufacturers of pumps and valves.

By sponsoring the ELLIS PhD Award, the Kühborth Stiftung aims

to support advances in artificial intelligence, one of the key technologies of the foreseeable future,
to promote academic excellence, and
to propagate the European idea.

ELLIS PhD Awardees

2024

Anastasiia Koloskova (EPFL École Polytechnique Fédérale de Lausanne, Switzerland)
‘Optimization Algorithms for Decentralized, Distributed and Collaborative Machine Learning’
Anastasiia has made groundbreaking progress towards the democratization of machine learning, through improving collaborative, distributed and federated learning. Her research empowers people and organizations both as contributors and as beneficiaries of machine learning systems, while providing significantly improved communication efficiency, asynchrony, privacy, in addition to a new fundamental understanding of these methods.

Luigi Gresele (Max Planck Institute for Intelligent Systems & University of Tübingen, Germany)
‘Learning Identifiable Representations: Independent Influences and Multiple Views’

Luigi offers a profound idea for ICA: rather than building on the classical approach of seeking independence between latent factors, he seeks causally independent mixing mechanisms. This establishes an elegant connection to causality, further expanded in his works on causal representation learning, as well as (in a suitable limit) to VAEs. Luigi’s oeuvre further contains original work on Simpson’s paradox in COVID-19, and a causal digital twin for vaccine allocation, rendering his research unique in both depth and breadth.

Runner Up: Jonathan Schwarz (University College London, UK)

‘Sparse Parameterisations for Efﬁcient Machine Learning Algorithms’

We celebrate Jonathan Richard Schwarz’s research on advancing efﬁcient machine learning through sparse parameterizations. His work constitutes a unique argument for an algorithmic principle while connecting a wide range of ideas under a shared paradigm. Jonathan’s thesis presents various empirically successful methods covering a wide range of problems. Finally, we also applaud his commitment to academic service, as evidenced by his active community involvement.

Learn more about the 2024 Awardees

2023

Pepa Atanasova (University of Copenhagen, Denmark)
Accountable and Explainable Methods for Complex Reasoning over Text'
Entrusting machines with the responsibility of performing complex reasoning tasks such as fact-checking makes it imperative that we understand how they arrive at their conclusions. The publications in Pepa's thesis collectively contribute to advancing the state of the art of accountable and transparent machine learning (ML) for complex reasoning over text.The methodological contributions of the thesis in the area of accountability enable the automated generation of challenge and adversarial datasets revealing important insights about models' flaws and capabilitites. In the area of explainability, the most noteworthy scientific outcomes include the proposed novel task of generating free-text explanations for fact checking predictions together with the first method for the task. Finally, the thesis contributes with explainability diagnostics, which are further optimised in the generated explanations, thus improving explanation quality. Read more here.

Robert Geirhos (University of Tübingen)
To err is human? A functional comparison of human and machine decision-making'
"Do machines see the world like humans?" - Robert Geirhos' interdisciplinary dissertation pioneered quantitative analysis methods for comparing human and machine perception. His research has challenged widespread assumptions about how machines see the world and led to the following insights: ML models recognize objects by their texture (rather than by shape like humans), they often exploit simple shortcuts instead of learning the intended solution, and they make very different errors compared to humans. Despite these differences, the gap between human and machine vision is steadily narrowing - a development that can be tracked using the Open Source benchmarks developed in this thesis. By stress-testing the limits and biases of machine learning models, Robert Geirhos' dissertation underscores the critical importance of understanding how machines perceive the world around us. Read more here.

Learn more about the 2023 Awardees

2022

Mathilde Caron (Facebook AI Research & Inria Grenoble)
’Self-Supervised Learning of Deep Visual Representations’
Mathilde Caron focused on creating learning machines that provide the ability to solve visual recognition tasks without relying on annotated data. The fundamental idea: Humans and many animals can see the world and understand it effortlessly, which gives hope that visual perception could be realised by computers and AI. More importantly, living beings acquire such an understanding of the visual world autonomously without the intervention of a supervisor explicitly telling them what is to be seen. This suggests that visual perception can be achieved without too much explicit human supervision, but simply by letting systems observe large amounts of visual inputs. Mathilde Caron tackles the problem of self-supervised learning which consists in training deep neural networks without using any human annotations. Read more here.

Taco Cohen (University of Amsterdam)
Equivariant convolutional networks’
Taco Cohen focused on improving the statistical efficiency of neural networks by enabling them to exploit the symmetries of learning problems. The underlying problem: While human beings and other intelligent animals have the ability to learn new skills and concepts very quickly, current machine learning systems that are able to acquire basic motor skills and perceptual capabilities require a very large amount of training data to do so. Because of this limited ability to generalise, the long tail of ML problems remains inaccessible. Taco Cohen explores ways to leverage symmetries to improve the ability of Convolutional Neural Networks (CNNs) and other kinds of neural networks to generalise from relatively small samples. Read more here.

Learn more about the 2022 Awardees.

2021

Thomas Kipf (University of Amsterdam, Netherlands)

Deep Learning with Graph-Structured Representations

Thomas Kipf's dissertation pioneered algorithms and neural architectures for learning with structured data in the form of networks or graphs, and for modeling interacting dynamical systems using graph-structured representations. In particular, his thesis has introduced influential graph neural network models, such as the graph convolutional network and the graph auto-encoder, which have found widespread application in areas such as drug discovery and (social) network analysis.

Arsha Nagrani (University of Oxford, United Kingdom)

Video Understanding using Multimodal Deep Learning

Arsha Nagrani's dissertation introduces multiple creative ways to use multi-modal data in machine learning. It explores cross-modal and self-supervised learning for important applications such as speaker recognition, action recognition, and text-to-video retrieval. It also contributes a data generation framework that has resulted in several large-scale audio-visual datasets. Both the scientific outcomes of the thesis and the datasets released have had a high impact in terms of citations and downloads.

Learn more about the 2021 Awardees

2020

Felix Berkenkamp (ETH Zurich, Switzerland)

Safe Exploration in Reinforcement Learning: Theory and Applications in Robotics

Felix Berkenkamp’s dissertation breaks new ground by pioneering provably safe reinforcement learning algorithms and demonstrating them on robotic platforms. His approaches use probabilistic models to cautiously learn about the system online, while eventually enabling increasingly performant behavior as supported by data. The dissertation elegantly integrates ideas from machine learning (such as concentration bounds for Gaussian process models, neural network policies) and control theory (such as Lyapunov stability and model predictive control).

Gül Varol (INRIA & Ecole Normale Supérieure Paris, France)

Learning Human Body and Human Action Representations from Visual Data

The thesis of Gül Varol covers a wide range of challenging and important problems in human analysis within computer vision. In particular, it defined the state of the art in action recognition from videos with the use of long-term temporal convolutions, and it contributed a synthetic data generation framework for advancing human body shape analysis. The scientific outcomes of the thesis have been followed up and extended by numerous groups. This has been supported by open sourcing of code and datasets.

Learn more about the 2020 Awardees

2019

Maria Barrett (University of Copenhagen, Denmark)

Improving Natural Language Processing with Human Data

The dissertation makes an important theoretical contribution to the field of NLP by showing that traces of human processing, particularly eye-tracking data, can be used to improve models for a range of language processing tasks. It represents impressive innovative work, makes significant novel contributions to various NLP tasks and each of the studies presents clear hypotheses and follows a well-justified methodology. (...) Thus, the thesis as a whole advances the state- of-the-art NLP research by suggesting what can be defined a new and promising modelling paradigm.

Wittawat Jitkrittum (Gatsby Unit, University College London, UK)

Kernel-Based Distribution Features for Statistical Tests and Bayesian Inference

Wittawat Jitkrittum's thesis addresses the problem of finding computationally efficient and informative features for representing and comparing distributions, while making as few assumptions as possible. The proposed two-sample test, independence test, and goodness-of-fit test are the first to combine consistency against all alternatives, adaptivity to maximise test power, efficiency (linear runtime), and informativeness (showing why a decision is made).

Alex Kendall (University of Cambridge, UK)

Geometry and Uncertainty in Deep Learning for Computer Vision

Alex Kendall’s dissertation introduced a number of exceptional new research contributions in the field of computer vision and deep learning. In particular, it advanced state-of-the-art in semantic segmentation, localisation, stereo vision and scene understanding using ideas from multi-view geometry and probabilistic modelling.

Diederik Kingma (University of Amsterdam, Netherlands)

Variational Inference and Deep Learning: A New Synthesis

Diederik proposed novel efficient solutions to the problems in variational (Bayesian) inference, generative modeling, representation learning, semi-supervised learning, and stochastic optimization. Among the proposed methods are the Adam optimization method, the Variational Autoencoder using the Reparameterization Trick, Inverse Autoregressive Flows, Weight Normalization and the Local Reparameterization Trick.

Anastasia Pentina (Institute of Science and Technology Austria, Austria)

Theoretical foundations of multi-task and lifelong learning

Anastasia Pentina contributed to expanding the understanding of potential benefits of information transfer in the multi-task and lifelong learning scenarios. She formally identified and analysed several scenarios of transfer learning. From the theoretical results she derived new principled multi-task and lifelong learning algorithms and illustrated their performance on various synthetic and real-world datasets.

Siyu Tang (Max Planck Institute for Informatics, Germany)

People detection and tracking in crowded scenes

Siyu Tang made significant contributions in the area of multi-person tracking and detection. She proposed novel and rigorous mathematical formulations based on the Minimum Cost Multicut models that perform well for real-word and challenging visual scenes and thus have defined the state-of-the-art in this area.

Learn more about the 2019 Awardees

ELLIS Newsletter

If you want to receive the ELLIS newsletter regularly via email, please subscribe here:

Imprint | Privacy Policy | Logos | Contact