Multitask Visual Representation Learning

Duy-Kien Nguyen (Ph.D. Student)

In order to build a comprehensive vision system at the human level, it is necessary for the system to be able to deal with multiple problems at different levels. Previously, researchers have focused on individual tasks separately with a specific network architecture design and many hand-crafted modules, which makes ...

3D Human Pose Estimation

Nadine Rüegg (Ph.D. Student)

I'm working at the intersection between Computer Vision and Machine Learning. Specifically, I focus on 3D human shape and pose estimation and have strong interests in unsupervised learning. I aim to find a good balance between labeling effort and performance.

3D Vision Meets Deep Learning

Songyou Peng (Ph.D. Student)

My research interests lie at the intersection of deep learning and computer vision, especially 3D vision. What could be the optimal 3D representation? How to effectively and efficiently combine deep learning with 3D vision tasks? How to tackle multiple 3D tasks together with less or no supervision? During the period...

A game theoretic framework for reinforcement learning and imitation learning

Luca Viano (Ph.D. Student)

Reinforcement Learning and Imitation Learning achieved impressive empirical results in recent years. Still, little is known about the theoretical properties of commonly used algorithms. This PhD project aims to use insights from the field of game theory and optimization to develop algorithms with mathematical guaran...

Accountability for AI

Florian Dorner (Ph.D. Student)

With the more and more widespread deployment of Al systems in the real world, they are likely going to have larger and larger impacts on society, including accidents, political destabilization and other negative externalities. To increase developers' accountability for negative societal impacts of their Al systems, ...

Advancements in 3D Scene Understanding: From Few-shot Segmentation to Multi-modal Perception

Zhaochong An (Ph.D. Student)

Accurate 3D scene understanding allows machines to recognize objects, perceive spatial relationships, infer scene semantics, and predict object behavior. This understanding is essential for many applications such as autonomous navigation, augmented and virtual reality, and robotics. One of the main challenges in dev...

Advancing generative models for 3D synthesis

Olga Grebenkova (Ph.D. Student)

The goal of this PhD project is to investigate the potential of generative models in 3D for creating high-quality, realistic objects suitable for various applications. While significant progress has been made in text-to-image synthesis using diffusion models trained on billions of image-text pairs, applying this tec...

Advancing SLAM through Deep Learning

Niclas Vödisch (Ph.D. Student)

A fundamental challenge in the field of mobile robotics is to determine the robot’s pose relative to its environment. Commonly, this is achieved by employing simultaneous localization and mapping (SLAM) systems. While classical methods are well studied and provide reliable solutions to the localization problem, how ...

AI Fairness in Federated Learning

Gergely Dániel Németh (Ph.D. Student)

Federated Learning is a rapidly emerging approach to solve privacy problems regarding Machine Learning applications. Fairness in Artificial Intelligence is also a field that gained awareness in recent years. At the intersection of the two fields, we can find new solutions for problems that exist in centralized machi...

Analyzing and Improving the Out-Of-Distribution Generalization Capability of Visual Foundation Models

Mohamed Afham (Ph.D. Student)

Technological advancements, fueling the abundance of online visual content, have enabled extensive pre-training of computer vision foundation models. In computer vision literature foundation models are referred to as the base models trained on large-scale data in a self-supervised or semi-supervised manner that can ...

Anytime Uncertainty in (Bayesian) Deep Learning

Metod Jazbec (Ph.D. Student)

The predictions generated by neural networks are fairly useless without a corresponding notion of confidence or uncertainty. Accordingly, methods for cheaply quantifying the uncertainty of NNs have received much attention. Yet, one issue that is often ignored is that our models operate under dynamic computation cons...

Artificial Scientific Discovery

Antonio Norelli (Ph.D. Student)

Explanations are the fuel of progress, the fundamental tool through which humans have increased their agency, earning more and more control over their future throughout history. So far, the production of these extraordinary symbolic sequences has been a unique prerogative of human scientists, but the formidable brea...

Automated Painting: Object Modeling, Trajectory Learning and Adaptation

Gabriele Tiboni (Ph.D. Student)

Spray painting is a task commonly performed by robots in industry. Automating this process offers advantages on the consistency of the results and repeatability over even the most skilled human operator, with the further benefit of limiting human exposure to hazardous environments. The goal of this thesis is to redu...

Bayesian deep learning for medical time series

Fedor Sergeev (Ph.D. Student)

Deep learning (DL) has seen tremendous growth in the last 20 years and enabled rapid advances in computer vision, natural language processing, and robotics. However, the direct application of DL methods to healthcare is challenging due to the nature of medical data. Creating specialized data sets for DL in medical a...

Bridging Theory and Algorithms for Lifelong Learning

Marco Ciccone (PostDoc)

One of the most ambitious goals of Artificial Intelligence is to provide an agent with the ability to continually learn in an open-ended scenario. Lifelong Learning, or Continual Learning (CL), explicitly focuses on non-stationary or changing environments, where an agent has to complete a series of partially related...

Causal Effect Estimation with Hidden Confounding and Aggregated Data

Yuchen Zhu (Ph.D. Student)

Causal effect estimation is ubiquitous in many social science and public health domains. In such domains treatment variables and outcome variables are often confounded, and sometimes the confounding is hidden, requiring methods capable of correcting for these. Moreover, in many social science domains the variables...

Causal inference under Gaussian Process

Zonghao Chen (Ph.D. Student)

The project aims to incorporate the reliable uncertainty estimate from Gaussian Processes (GP) into causality inference. Recent work on causality inference based on kernel methods have demonstrated both theoretical guarantees and good empirical performances. Therefore, the aim of this project to extend from kernel m...

Causal modelling and applications to domain adaptation

Philiip Boeken (Ph.D. Student)

In this project, we investigate multiple aspects of causal modelling. We have investigated how causal models can be used to correct for selection bias and missing response when learning a regression model. In a follow-up project, we have investigated how the practical use of predictions can affect the value of the v...

Causal Representations for Reinforcement Learning

Yashas Annadani (Ph.D. Student)

Systematic generalisation beyond the independent and identically distributed (IID) setting is crucial for learning agents interacting with the environment in the real-world. Improvements with regards to generalisation has been seen in supervised learning with large language and vision models and out-of-distribution ...

Causality of Enhanced Model Interpretability

Amir-Hossein Karimi (Ph.D. Student)

As machine learning is increasingly used to inform decision-making in consequential real-world settings (e.g., pre-trial bail, loan approval, or prescribing life-altering medication), it becomes important to explain how the system arrived at its decision, and also suggest actions to achieve a favorable decision. My ...

Combining Probabilistic Inference and Deep Learning

James Urquhart Allingham (Ph.D. Student)

Deep learning is an extremely successful machine learning paradigm that has been the driving force behind the renewed excitement around machine learning over the last seven or so years since AlexNet won the ImageNet LSVRC competition. However, despite the successes of deep learning, it is not without its flaws. One ...

Computational Rationality in AI-Assisted Decision Making

Yifan Zhu (Ph.D. Student)

Effective and smooth human-AI collaboration in decision-making requires the AI assistant to accurately comprehend human intentions by observing behaviour. While much of the research in this field emphasizes advanced AI techniques for inverse modelling and obtains promising results, it often oversimplifies human beha...

Continual Learning in Computer Vision

Sebastian Dziadzio (Ph.D. Student)

The dominant deep learning setting involves training a model offline on a fixed dataset before deploying it in the real world. Implicit in this approach is an important and restrictive assumption: both the training data and future inputs are assumed to be IID samples from a single stationary distribution. In reality...

Continual Open-set Object Detection

Yaoyao Liu (Ph.D. Student)

Most object detection methods require abundant labeled training samples per category. However, we often receive a continuous temporal stream of visual data with many unknown categories in real-world applications. For example, a robotic learning system needs to deal with new objects observed in the environments but c...

Continually Learning to Detect Distribution Shifts

Mona Schirmer (Ph.D. Student)

Autonomous vehicles are continuously collecting data, and the system must decide which observations are sufficiently interesting enough to save for future use. In settings like this, one cannot use a static model to determine what constitutes an anomaly. For instance, when the car first encounters snow, this is like...

Cost-efficient learning via weak annotators

Hao Qiu (Ph.D. Student)

In this project, we plan to design and analyze new algorithms for sequentially classifying a stream of data points based on a set of costly, noisy, and potentially malicious annotators. Assuming the algorithm obtains noisy labels by adaptively querying selected annotators, we are interested in studying trade-offs be...

Cross-Domain 3D Visual Learning

Antonio Alliegro (Ph.D. Student)

Depth cameras and LiDAR sensors are important tools for agents that need to perceive the world and interact with it, thus 3D data managing and learning algorithms are quickly becoming essential. Deep learning has achieved a significant leap for 2D computer vision, and for 3D applications is currently demonstrating g...

Cross-Lingual Document Summarization

Yifu Qiu (Ph.D. Student)

Automatic document summarization is an area in Natural Language Processing focused on taking input documents and outputting a concise summary of them. Such summaries allow potential readers to sift through a more significant amount of information in a shorter time and focus on the most critical content. Recent work ...

Data and Compute Efficient Adaptation of Large Pretrained Models

Konstantin Dobler (Ph.D. Student)

Publicly released checkpoints of large pretrained models are quickly becoming ubiquitous. These checkpoints are an important resource in practice due to the large amount of compute invested in them. However, for most real-world use cases, these checkpoints will have to be adapted to obtain the best performance. In m...

Debiasing Multilingual Representations by Discovering Shared Bias Components in Multiple Languages

Shun Shao (Ph.D. Student)

Modern NLP models that learn from tremendous data are extremely powerful. However, they also tend to associate certain gender, race, or religion with the label, and make unfair and biased predictions. Multilingual debiasing is a promising topic that tries to learn a debiasing model from multiple languages, and debia...

Design and development of computational methods to facilitate the analysis of single-cell multi-omics and single-cell perturbational data

Amir Ali Moinfar (Ph.D. Student)

Single-cell omics allow us to study biological processes at single-cell resolution. The applications of single-cell omic studies range from understanding cellular processes to drug design. We need to design and develop computational models that can handle the complexity and magnitude of data generated by these studi...

Dynamically Evolving Deep Architectures

Rupert Mitchell (Ph.D. Student)

My research focuses on the modification of deep architectures live during training. In contrast to the relatively well explored area of the pure removal of parts of an architecture (i.e. pruning), I am particularly interested in combining this with the addition of parts as well, for example increasing the width or d...

Efficient Deep Learning for Sets

David Zhang (Ph.D. Student)

Sets help us organize our understanding of the real world: complex images can be simplified as sets of objects, arbitrary 3D shapes can be described by sets of coordinates, and graphs (a tuple of two sets) can model all relational structures between objects. My research focuses on inferring sets with deep learning a...

Efficient Learning and Rendering of Neural Representation for Mobile Devices

Stefano Esposito (Ph.D. Student)

Neural Radiance Fields (NeRFs) can create images of 3D scenes from new perspectives, but real-time rendering is hard. Some methods for real-time rendering exist, but they have trade-offs and require high-end GPUs. This study aims to develop training procedures and rendering algorithms for neural implicit surfaces su...

Efficient Video Representation Learning

Mohammadreza Salehi (Ph.D. Student)

In recent years, representation learning has made significant strides, enabling self-supervised learning methods to achieve comparable performance to supervised approaches. However, these models typically require extensive training datasets and substantial computational resources to achieve their optimal performance...

Enabling complex computer vision tasks on the edge: applications to semantic segmentation and beyond

Claudia Cuttano (Ph.D. Student)

Most of the research in Deep Learning for Computer Vision is dedicated to the development of novel, more and more complex and overparametrized architectures which increase state of the art accuracy in solving specific tasks. However, this approach lays its foundations on the strong assumption that it is always possi...

Essentials for few-shot learning on images

Jelena Bratulic (Ph.D. Student)

Few-shot learning strives to imitate the ability of humans to learn how to recognize an object after only having seen a small number of samples of the new object. This is achieved by leveraging the rich world model that humans have learned over their life span to explain novel patterns as variations of known pattern...

Explainable De-biasing in Learning from Interactions

Maria Heuss (Ph.D. Student)

Explainability and Fairness are two topics of great importance when working with machine learning models that might not be intrinsically transparent. In this project we are looking a these topics in the context of ranking systems. The nature of user behavior, when interacting with ranking systems, is inherently bias...

Explainable Multimodal Representation Learning

Karsten Roth (Ph.D. Student)

Human perception, but also interaction, is inherently multimodal, i.e. visual and audio information get processed together and the emerging knowledge is communicated through some form of language. In this project, our aim is to learn multimodal representations using artificial neural networks. Our application domain...

Facial Expression Applications and Analysis using Micro- and Macro-Expressions

Mengting Wei (Ph.D. Student)

Facial expressions (FE) are a kind of very intuitive and powerful signal to disclose people's emotions. With the development of deep learning methods, many efforts have been committed into predicting emotional states via facial expressions. In this research, we'll push this envelop further. First, we adopt FE as an ...

Fair Machine Learning Systems

Benedikt Höltgen (Ph.D. Student)

As consequential decisions affecting people's lives are increasingly being delegated to Machine Learning systems, society now needs to grapple with questions of algorithmic fairness. While this is still a fairly new area of research, similar questions have long been discussed both informally and mathematically in ne...

Federated Semantic Segmentation architectures on IoT devices using Neural Architecture Search

Shyam Nandan Rai (Ph.D. Student)

Scene understanding is one of the key components for robotic navigation and self-driving application. Semantic segmentation gives detailed information about a scene as it involves pixel-wise dense labeling. However, semantic segmentation models are challenging to deploy on resource constraint devices due to high com...

Few-shot Learning with Pretrained Language Models

Abdullatif Köksal (Ph.D. Student)

Large pretrained language models (PLMs) perform well on many NLP tasks without supervised training. This phenomenon is best demonstrated in GPT-3 by its translation capability from different languages to English with just 64 contextual examples. On the other hand, contextual examples enable PLMs to work with multipl...

Flow Matching for domain translation

Luca Eyring (Ph.D. Student)

Continuous Normalizing Flows (CNFs) are a family of continuous-time deep generative models. Flow Matching (FM) has been proposed for simulation-free training of CNFs based on conditional probability paths between a source and target distribution. This enables training based upon a simple regressions objective simila...

Generalizeable Video Representation Learning

Michael Dorkenwald (Ph.D. Student)

Video data is a treasure trove for AI models, providing a lens to the intricate dynamics and mechanisms that define our world. The key to unlocking the tremendous amount of data available on the web is to bypass the laborious and expensive process of annotating each video. Yet, extracting knowledge and understanding...

Generative Models and Stochastic Processes

Cameron Stewart (Ph.D. Student)

This project will focus on the further development and analysis of generative models. Examples of such models include Generative Adversarial Networks (GANs), energy-based models, normalising flows, diffusion models, and score-based generative models. Wasserstein gradient flows of various metrics applied to the train...

Generative Models for Virtual Humans

Anton Zubekhin (Ph.D. Student)

With the recent advancements in generative modelling, it is now possible to create high-quality images and videos from scratch with greater control. Simultaneously, the field of 3D reconstruction and novel view synthesis of human bodies has been rapidly developing. Despite these advancements, many challenges remain ...

Generative Models with Fluid Dynamics

Chenyu Zhang (Ph.D. Student)

In generative models, the challenge lies in sampling and forming images from an original distribution, which can be reframed as a problem of transforming one distribution into another. To address this, some generative models, including normalizing flow, score matching models, and the recent flow matching methods are...

Geometric and statistical software engineering tools in array-centric programming

Joanna Sliwa (Ph.D. Student)

In the course of training neural networks, the evaluation of the training process is often solely based on the determined loss. Prior research has attempted to incorporate software engineering tools to enable an assessment beyond just the gradient, perform it efficiently, and offer a more comprehensible evaluation o...

Geometrie deep learning for shape correspondence, graph data and quantum chemistry

Christian Koke (Ph.D. Student)

A classic task in Computer vision is finding correspondences between geometric shapes, often represented via triangle graphs. Algorithms achieving state of the art performance on this task built on two key ingredients: A) The generation of informative local (geometry-capturing) features. B) A global transfer schem...

Grounding symbols in observations via causal, object centric world-models

Anson Lei (Ph.D. Student)

Symbol grounding is a key part of classical AI and harbours significant potential in the ability to perform symbolic and causal reasoning on real-world data. In contrast, machine learning currently operates on raw sensor data and typically without consideration of any symbolic structure in the data. The recent emerg...

Harmonic Analysis for Stochastic Neural Networks

Adeel Pervez (Ph.D. Student)

In this work we explore harmonic analytic techniques to improve the learning capabilities of stochastic neural network models. Stochastic neural networks are widely used in probabilistic modeling for unsupervised and representation learning. Such models are often limited by problems of instability and by the type of...

Holistic Approaches for Dynamic Scene Understanding

Maximilian Luz (Ph.D. Student)

Perception—and, in a more direct sense, dynamic scene understanding—is fundamental to the success of autonomous robots. Adequately capturing and representing environmental complexity and dynamics in a reliable and robust manner is critical to subsequent tasks such as (local) navigation and planning, especially in en...

Human Modeling via Reinforcement Learning and Computer Vision

Berat Mert Albaba (Ph.D. Student)

Modeling and imitating human behaviors, actions and decisions from visual data are crucial for the future of AI, yet problematic due to the complexity of humans. In this project, the goal is proposing approaches to model and imitate humans in a generative manner. Thanks to its well adaptation ability to new scenario...

Human-centric Explainable AI

Julien Colin (Ph.D. Student)

The goal of Explainable AI (XAI) is to develop methods that explain to humans how AI models behave so humans (including non-experts) can understand the basis for the algorithm’s decisions. Within the computer vision literature, the most widely used explainable AI methods are attribution methods. Recent works on huma...

Implicit Generation of Graph Structures

Gbètondji Dovonon (Ph.D. Student)

Graphs are natural representations for data in a wide array of applications, ranging from materials science to algorithm design. Often however, these graphs are not available in data: e.g., a potential library of ideal catalysts may not have known synthesis graphs required to construct them. In this project our goal...

Incremental Learning in Semantic Segmentation

Fabio Cermelli (Ph.D. Student)

Semantic Segmentation is a crucial task in computer vision that consists of predicting for each pixel of an image its semantic meaning, i.e. pixel-wise classification. This task is crucial for many applications, ranging from self-driving cars to autonomous manufacturing. In the last years, thanks to the rise of deep...

Inductive and Semantic Priors for Categorization in Deep Learning

Tejaswi Kasarla (Ph.D. Student)

An inductive bias of a learning algorithm describes a set of assumptions about the target function independent of training data. Inductive biases play a vital role in the design of machine learning algorithms, consider for example inductive biases for image structures ( e.g., the convolution operator), symmetries (e...

Interactive Multimodal Learning

Haau-Sing Li (Ph.D. Student)

Multimodal learning contains a set of challenging tasks. It requires a deep understanding of all relevant modalities (language, vision, software program) and of the relationships between them. Furthermore, the lack of high-quality datasets increases the difficulty of related tasks. In this project, we will work with...

Interpretable Machine Learning for Hydrological Drought Understanding

Georgios Blougouras (Ph.D. Student)

Climate change and direct human activities are profoundly influencing the water cycle through complex interactions, couplings, and feedback mechanisms. An acceleration of the terrestrial component of the water cycle implies increased frequency of hydrological droughts, whose characteristics are also altered in respo...

Interpretable Representation Learning using Generative Models

Silpa Vadakkeeveetil Sreelatha (Ph.D. Student)

Generative models have witnessed significant performance improvements in image synthesis over the last decade with the introduction of the generative adversarial network (GAN), variational autoencoder (VAE), and Diffusion models. Extensive research has been carried out to demonstrate their utility in applications su...

Interpreting NLP models through the lens of cognition and linguistics

Michael Hanna (Ph.D. Student)

Transformer-based models have spurred unprecedented increases in performance on NLP tasks. As these models' performance begins to approach (or even surpass) that of humans, the question of how they achieve this performance becomes crucial. Do these models acquire the same linguistic knowledge, and learn to perform t...

Intrinsic and Invariant Image Decomposition

Xiaoyan Xing (Ph.D. Student)

Intrinsic and invariant information are fundamental components of the image formation process. In fact, the image formation heavily depends on the geometry of the scene, object material, and illumination. To decompose the image into its constituent parts, well-awareness of the geometry information in the scene and t...

Investigating and preventing harmful cultural implications of AI in social media

Piera Riccio (Ph.D. Student)

Social media platforms are reshaping the flow of information in our society, establishing new paradigms of mass communication. Because of their ubiquitous nature, they constantly influence our daily choices, having a widespread impact on different facets of our culture. On such platforms, AI algorithms are massively...

Joint Sentence and Word Alignment

Peiqin Lin (Ph.D. Student)

Word alignment is generally decomposed into two subproblems: first align sentences, then align words on the assumption that only words in aligned sentences can be aligned. This decomposition works well for many genres of parallel corpora, in particular, for parliamentary proceedings and for legal and business text. ...

Latent Communication in Artificial Neural Networks

Luca Moschella (Ph.D. Student)

As neural networks permeate various scientific and industrial domains, understanding the universality and reusability of their representations becomes crucial. At their core, these networks create intermediate encodings or representations of input data and subsequently leverage them to perform specific downstream ta...

Learning compact and efficient feature representations

Zhuo Su (Ph.D. Student)

Energy efficient sensing and computing is vital at all levels, from the smallest sensor like the chip to ultra high performance processors and systems like the cloud, especially in the post Moore's Law era. Energy efficient AI enables AI to move beyond the cloud and to reach the edge, which is critical to the progre...

Learning from Context with Generative Models

Ivona Najdenkoska (Ph.D. Student)

Empowered by the large-scale pre-training on huge webscraped datasets, large language models (LLMs) have witnessed major advancements in recent years. These models demonstrate fascinating emergent abilities, particularly learning from a few examples presented in prompt with no gradient updates, also known as in-cont...

Learning from multiple distributions

Maximilian Beck (Ph.D. Student)

Traditional deep learning methods perform well in single-domain settings where a vast amount of data is available, but fail in applications where labeled data is limited and comes from different domains. This is for example the case when industrial sensor systems need adjustments to new operating conditions or when ...

Learning methods for geometry and appearance reconstruction

Božidar Antić (Ph.D. Student)

As XR technologies are gaining more traction, estimating detailed geometry and appearance of scenes and humans is becoming a crucial task in order to have realistic and interactive virtual experiences. The majority of today's methods focus only on reconstructing geometry and simple textures. However, estimation of u...

Learning on Graph Problems in Biology

Paolo Pellizzoni (Ph.D. Student)

Developing machine learning methods for graph-structured data is a theme of paramount importance in bioinformatics. Indeed, graphs are the structure of choice for representing complex objects such as molecules and proteins, which have a central role in many biological problems, as well as for capturing the interacti...

Learning Robot Perception and Localization with Limited Supervision

Julia Hindel (Ph.D. Student)

Comprehensive scene understanding and state estimation are pivotal for achieving reliable robot autonomy in human-centred environments. Yet, the availability of reliable training data undermines the robustness - and thus safety - of such systems, especially in edge cases not or barely covered. This project focuses o...

Learning to solve Differential Equations with Uncertainty

Emilia Magnani (Ph.D. Student)

Learning differential equations is an emerging research theme in machine learning. Differential equations are interesting because they offer a powerful language for dynamical relationships between variables and a mechanism for reduction for structured models, which is especially relevant in science. The project will...

Learning with Geometry on real world data

Marco Pegoraro (Ph.D. Student)

This PhD project delves into harnessing the intrinsic geometric properties of data to amplify learning tasks, with a primary focus on biological data. Traditional Euclidean models often prove inadequate in capturing the intricate structures inherent in real-world data. Consequently, we delve into nuanced spaces th...

Learning-based 3D Approach for Egocentric Vision

Weirong Chen (Ph.D. Student)

Egocentric vision, which involves understanding and interpreting visual data from a first-person perspective, is crucial for various applications, including augmented reality, robotics, and assistive technologies. This project seeks to advance the current state of visual foundation models while addressing challenges...

lmproving Image Synthesis and Manipulation Control with GANs

Edgar Schönfeld (Ph.D. Student)

Among the major challenges of generative adversarial networks (GANs) remain the ability to create images that are indistinguishable from real images and the ability to depict precisely the content specified by the user. To this end, this work proposes methods to improve synthesis quality and maximize control over th...

Long-Horizon Planning and Reasoning for Intelligent Robotic Systems

Imen Mahdi (Ph.D. Student)

Vision Language Action (VLA) models have been proposed to enable robots to execute a variety of tasks by integrating perception, language understanding, and action planning. Large VLAs trained on a diverse set of tasks have been shown to generalize well to new tasks and environments, achieving high precision and eff...

Machine learning approach for multi-scale genomics problems

Olga Mineeva (Ph.D. Student)

In the age of rapid growth of available biological sequencing data enabled by the recent advances in sequencing technologies, there is an opportunity to answer biological and health-related questions at a more detailed level. At the same time, the amount of data allows the use of sophisticated methods, such that dee...

Machine learning for improving climate models

Fernando Iglesias-Suarez (PostDoc)

Earth system and climate models are fundamental to understanding and projecting climate change. Although they have improved significantly over the last decades, considerable biases compared to observations and uncertainties in their projections still remain. We will take a new approach by harvesting output from high...

Machine Learning for Spatial Omics in Perturbed Systems

Florin Walter (Ph.D. Student)

Spatial omics technologies enable the study of omics modalities such as the genome, transcriptome, proteome or epigenome in their spatial context, thereby allowing a deeper understanding of cell function, cell communication or tissue development. Spatial technologies based on fluorescence imaging generate single-mol...

Machine learning methods for graph-structured data

Leslie O'Bray (Ph.D. Student)

Alongside the rise of machine learning methods has been an increase in graph-structured data, which captures rich and complex relationships in the data. Since many machine learning methods do not work out-of-the box with graph-structured data, specific research and focus is necessary to develop such methods to enabl...

Mathematical Foundations for Deep Learning

Zhenyu Zhu (Ph.D. Student)

Despite the extraordinary success of deep learning, its emerging weaknesses such as robustness, generalization, and bias, demand an ever closer attention. Unfortunately, many of the existing theories were developed for low capacity models and therefore do not account for the impressive scaling properties of deep lea...

Methods for Robust Feature Learning

Sindy Löwe (Ph.D. Student)

The representation of objects and their relations lies at the center of our ability to reason and to adapt to new environments. In my PhD, I want to develop algorithms that learn object-centric representations from low-level perceptual inputs in an unsupervised fashion.

Model-based reinforcement learning and planning under uncertainty

Maris Galesloot (Ph.D. Student)

Reinforcement learning (RL) has recently seen increasing use due to its ability to scale decision-making under uncertainty to high-dimensional environments. To train RL agents, a simulator or part of a model specification is often used. However, the knowledge, in the form of explicit modelling of these simulators of...

Modeling Impacts of Extreme Events Using Reservoir Computing

Francesco Martinuzzi (Ph.D. Student)

Given their nature, extreme events are difficult to model for most Machine Learning algorithms when the data is not abundant. More specifically, daily Earth Observations from satellite data back only twenty years, making the data driven approach to extremes more complex. My goal is to investigate the ability of a re...

Modelling Clothed Humans via Neural Implicit Representations

Shaofei Wang (Ph.D. Student)