Current neural networks remain large and not fully understood. My research links network compression with model interpretability. This natural connection aims at extracting relevant information from the network while discarding the redundant parameters. This goal of reducing neural networks not only makes models faster, more practical and energy-efficient but also aims to analyze and interpret the relevant features that make up a smaller model. We look at this problem from various perspectives, ranging from probabilistic to game-theoretical approaches. My further interests lie in data generation and differential privacy.
I am interested in understanding deep learning from a mathematical perspective, using mostly differential geometry and group theory, as well as in deriving new algorithms based on this understanding.
My research focuses on social physical human robot interaction. Specifically I am interested in enabling a robot to give satisfying hugs to humans. By developing a custom designed robot for this particular kind of interaction, and integrating elements of computer vision and machine learning, I hope to enable people to send customised hugs to each other and enable the robot to work as a diagnostic tools.
To create robots that can solve complex tasks, we need them to understand how their environment works and how it is affected by their actions. I’m trying to solve this by combining simulation, perception and learning.
My research interest lies in computer vision and graphics. Specifically I am interested in 3D deep generative models, with a goal to build learning-based approaches capable of generating photo-realistic simulations of human activity. The resulting data could be useful in the context of training methods for human-centric perception tasks.
I am interested in modelling the human body, its motion and interaction with objects and scenes, which could then be used to generate new human-like actions and behavior for virtual agents.
My research interests lie in the broad field of computer vision and, more specifically, are currently focused on learning structured scene representations. I'm interested to learn about ways in which humans perceive their environment and how to use these ideas in combination with deep learning methods. Moreover, I want to explore options of combining this work with other tasks like segmentation or 3D reconstruction.
My primary research is at the intersection between Computer Vision, Computer Graphics and Machine Learning, with a focus on understanding 3-dimensional objects using deep learning technique. Currently, I am especially interested in the modeling and analysis of human faces and bodies.
Generally, in my research, I am interested in problems at the intersection of machine learning and domain sciences (in particular, astrophysics). More specifically, I want to investigate how we can incorporate and make use of existing scientific domain knowledge about a given problem to build better machine learning models. One specific project that I am currently working on aims to develop new ML-based post-processing algorithms for high-contrast imaging of extrasolar planets.
Bio-inspiration has become an established field in material and computer science as well as robotics. By examining biological characteristics and behaviors we can improve old technologies, develop new ones and combine them with existing methods. But not only technology benefits from the interaction between robotics and biology. Using robotic methods could help to proof biological hypotheses in e.g. morphology and locomotion. So what can we learn from and about the locomotion of spiders?
My research focuses on probabilistic machine learning and data science. Currently, I am particularly interested in approximate inference for neural networks and biomedical applications. I want to design algorithms that can incorporate prior knowledge, quantify uncertainty, and automatically select the most likely model given data. Especially in the context of neural networks, these problems remain challenging. I have previously worked on methods for time series, e.g., mobility and political data.
I am generally interested in human behavior, how intelligence is formed in humans, and how it can be replicated in machines. As a first year PhD student, I am exploring AI methods that are interpretable and explainable, with a focus on probabilistic inference on the organization of entities, or learning compositional and causal generative processes. My research has a broad impact, potentially in three areas: (1) more adoption of machine learning methods in critical social, economic, and public health domains due to the added trust gained from explainable methods; (2) better model selection strategies that point out hazardous practices with the abundance of data, including data leakage of irrelevant factors; (3) state-of-the-art models with improved fidelity/accuracy as we are able to conduct stage-wise learning of abstract objects and concepts and then complex relationships.
I am interested in computer vision and machine learning with a focus on human motion understanding.
In my research, I focus on enforcing desirable properties to the solution of learning algorithms, such as incorporating human beliefs, natural constraints, and causal structures. This translates to faster, more accurate, and more flexible models, which directly relates to real-world impact. I tackle this challenge on three sides: (i) I work on efficient constrained optimization algorithms that guarantee structural properties, provably converge and scale. (ii) I apply our novel theoretical insights to the problem of approximate inference with the goal of making it structured, efficient and accurate. (iii) I work towards learning algorithms that can spontaneously discover the natural structure present in the data, infer causal relations, and are useful for arbitrary downstream tasks.
I am interested in developing robust and interpretable Machine Learning methods for real-world problems, in particular, that arise in Healthcare and Genomics.
After a Bachelor's in Physics and Master's in "Computational Science & Engineering" at EPFL, I worked for Hoffmann-La Roche for roughly two years. My PhD, supervised by Bernhard Schölkopf and Gunnar Rätsch, will focus on causal inference, representation learning and applications to biological datasets. I previously worked on Bayesian optimization, variational inference and Gaussian processes. The beauty of probabilistic modelling thrills me.
I am interested in robots that are not only moving in but actively impacting their surroundings. My main field of research lies in the intersection of control, perception, and machine learning, intending to enable mobile robots to perform complex tasks. I am mainly working in the field of construction robotics, specifically on autonomous excavation and manipulation with a walking excavator.
My main research interests are out-of-sample generalization and causality in deep learning.
My research interests revolve around Computer Vision and Machine Learning and I am particularly interested in the development of methods capable of describing semantic content.
I have a broad interest in machine learning, including variational methods, causal inference and probabilistic programming. In the future, I hope to design adaptive machines that leverage prior knowledge to effectively learn in domains where traditional approaches still fail today.
My general research interest lies in computer vision and machine learning. More specifically, I am interested in how to apply deep learning to 3D vision problems.
In order to create intelligent machines, we should endow them with features connecting areas like machine learning and optimal control. Reinforcement learning lies at the intersection between these two areas and it will be the focus of my PhD, with a particular interest on model-based approaches and trajectory optimization.
My research interests include, but are not limited to, Probabilistic Learning and Network Science, as well as connected fields. In particular, I aim at understanding how current probabilistic models can be improved upon, both on a representation and training level. I am also fascinated by how different ideas and concepts from within and outside ML interpolate in interesting and novel developments. Therefore, I strive to keep a broader view on theoretical and practical insights originating from different fields.
My research focuses on interaction between humans and their environment, 3D pose and tracking. I hope by combining computer vision and machine learning methods, a holistic understanding of video scenes can be obtained.
My research focus lies at the interplay between control theory and machine learning. I try to make use of both techniques to build probably safe learning agents.
My research focuses on leveraging machine learning methods to model and control time continuous dynamical systems. Inspired by traditional scientific parametric model building, I am working on developing novel strategies for reinforcement learning, system identification and control.
My research interests lie in the intersection of unsupervised structured representations learning, dynamics learning and using both for model-based reinforcement learning. I find autonomous learning of environment representations that are modular, independently predictable and controllable as an important step towards efficient and accurate control for many real-life applications.
My research is concerned with how robots can learn to act in complex environments based on sensory information. In pursuing this I am interested in a wide range of reinforcement learning techniques and how they can potentially be applied in domains where the amount of training data is very limited.
I am interested in feature learning with applications in biomedicine, in particular in models that jointly learn representations either of the underlying biological systems or which can be useful for improving clinical practice.
I work in rehabilitation robotics and especially on patient-robot interaction with upper arm exoskeletons. We are developing human-machine learning algorithms to innovate robotic arm rehabilitation therapy and improve therapy outcomes for patients.
My research mainly focuses on deep learning, structured learning and optimization for big data. I study the theoretical and practical behaviors of learning systems and optimization algorithms, as well as their interactions while working with both large as well as high-dimensional data.
My research focuses on scene interpretation of remote sensing data at large scale. More specifically, I investigate the use of Machine Learning, mainly deep learning approaches, to model physical quantities like for example biomass from satellite data. Furthermore, I’m interested in time series analysis and change detection in the context of deforestation in tropical rainforests.
My research field is Computer Vision, or more specifically dense correspondence search for multi-view depth and flow reconstruction. I am currently working on volumetric particle flow reconstruction from image sequences of multiple cameras in order to estimate the 3D velocity field of fluids.
I work on the intersection between 3D computer vision and graphics. Specifically, I am interested in building learning-based models for non-rigid 3D shape analysis and generation.
My research interests focus on Bayesian Optimization applied to automatic tuning of feedback controllers for robots.
My long-term scientific goal is to solve fundamental problems in AI and Control. My research blends machine learning, control theory, statistical mechanics, and computational neuroscience towards that goal. My recent works have been on score matching, generative adversarial networks, dynamical systems, causal system identification, active learning, reinforcement learning, statistical learning theory, and logic. Previous to Ph.D., I worked on independent component analysis, optimization on Riemannian manifolds, Bayesian model selection, machine vision for medical applications, experimental neuroscience, and robotics.
My research focuses on the question of how users interact with dynamic systems. I want to understand how a user provides feedback and how the controller can learn and automatically adjust its control objective to match user-specific demands.
Can we build artificial systems which learn to understand and reason with natural language in the same effortless ways as ourselves? How is thinking realized in the brain? My main research interest is to build biologically and cognitively inspired machine learning models for Natural Language Processing, which can be easily benchmarked, and perform well on multiple real-world tasks.
My research interest is in 3D reconstruction and scene understanding/interpretation. I aim to develop systems and algorithms to create high-fidelity 3D models of physical objects and environments. I have build depth panoramas for semantic extraction in indoor settings and work on efficient semantic 3D modelling for large-scale urban areas.
My research focuses on Machine Learning methods over aerial and terrestrial images from sources such as satellite (optical and radar), drone and social media images. I am mostly interested in developing specific methods that can be used for environmental and humanitarian purposes taking advantage of the particulairties of hyperspectral and radar imagery.
My research focuses on deep generative models including approaches that apply high-level vision on low-level tasks such as super-resolution and more general models such as adversarial networks.
My research interests lie in developing and applying clean, mathematically sound methods to model the behavior of complex relations in high-dimensional data. More specifically, I am interested in learning the physiological states and trajectories of patients from their medical data.
My current research focuses on processing and analysis algorithms for data of different origin (RGB and depth images, image sequences, point clouds). I am also interested in the methods capable of working with other types of data such as time series and data without spatial dependency. I think, that the intersection of machine learning (deep) and domain specific applications allows to pose additional constraints for the model and extract the information hidden in the data more efficiently. That’s why generative models are also of a high interest for me. On the other hand, I always wonder if there exists the algorithm which can be agnostic to the problem and still be successful (sort of a universal soldier). I personally think that reinforcement learning is currently the most appealing technique for doing it.
I am interested in the the questions of how robots can safely explore unknown environments and how they can improve their controllers without failures. In particular, this currently involves topics around Bayesian optimization, robust control, and reinforcement learning.
I have broad interests in topics in learning theory, model/algorithm selection, probabilistic models, (large-scale) optimization & computer vision.
Daniel passed away in December 2018. His research combined principles from Artificial Intelligence, such as Machine Learning, Natural Language Processing and Computer Vision, with concepts from psychology and computer graphics. He was most interested in understanding human intelligence by developing new techniques that leverage tools from these disciplines. In other words, what makes human perception different from other animals? Is it language? Emotion? Both?
My interests lie in the statistical-computational trade-offs of large scale learning. The outcome of my research would be a learning method that considers the computational limits and adapts itself to obtain the best statistical accuracy.
I'm working on machine learning, causal inference and time series analysis for multi-agent and economic decision making problems.
My research interests are in control and haptic interaction between human and robots in sports and upper limb rehabilitation. My focus is on the design and evaluation of strategies to autonomously adapt robot control and to autonomously configure robot assisted exercises.
I work on the development of novel magnetic microrobot control and fabrication for micro-manufacturing. My current work includes manipulation of sub-mm objects in aqueous environments by capillary attachment and the control of magnetic microrobots with six-degrees-of-freedom.
Many combinatorial algorithms gradually reduce the disorder of inputs while running and finally give a precise output. This information "extraction" power of an algorithm is often related to its robustness on noisy inputs. My research aims at formalizing this insights and combines information theory, combinatorics and statistical physics.
I am interested in estimating 3D scene representations from multi-view video sequences. In particular, I focus on combining semantic segmentation, object detection and classification with 3D reconstruction using efficient inference methods.
In broad terms I am interested in how nature solves optimality problems, in particular with regards to dynamics and locomotion. I am currently studying how changes in morphology affect learning and control in both animals and robots.
My research deals with autonomous motion generation for humanoid robots that locomote and interact with their environment through contact interaction.
Magnetic manipulation can control multi-scale robots and the tracking method is as important as a control system. To expand the capabilities of manipulation systems, I am investigating three-dimensional tracking methods using projective x-ray imaging technique and digital holography.
My research focuses on dense 3D modelling from images. The goal is to not only reconstruct the geometry but at the same time also reason about the semantic classes and the objects that are present in the scene. I'm also interested in real-time robotics applications of dense 3D reconstruction.
I work at the intersection of machine learning, computer vision and computer graphics. Specifically, I am interested in devising new machine learning techniques for better inference in computer vision models.
I work on data-driven learning for dexterous robot manipulation. Specifically: object detection at speed, reactive robotic perception, bootstrapping data-driven methods for grasp planning, leveraging physics simulation, and exploiting sensory experience for decision making
I investigate the use of Gaussian processes in dual control -- the task of efficiently identifying the relevant parts of a dynamic system while simultaneously controlling it.
I work in computer vision, specifically on visual recognition and tracking of human bodies. I am interested in efficient non-parametric context models and their applications to spatial, temporal and semantic dependencies in the human body.
My research uses ideas from causal inference to improve on existing statistical inference procedures. This includes methods for finding significant causal variables in time series data, extensions to blind source separation under the presence of confounding as well as methods for structure identification of dynamical systems.
I work on robot-assisted Palaeoanthropology. I analyze stone tool artifacts from archaeological sites by microscopy regarding traces of wear and compare this with wear pattern created on stone tool replicas by hypothetical tasks. The use of robots allows to control parameters of the task and to create large reference collections of wear pattern easily.
I am interested in understanding the fundamental principles of robot locomotion and manipulation, how to optimally use its sensory information, and developing planning and control strategies that can leverage the available information to create truly robust and adaptive behaviors
The goal of my research is to compute and physically generate complex three-dimensional acoustic holograms. To achieve this I want to adapt existing algorithms and combine them with new techniques from machine learning. Furthermore, I'm interested in the physical realization of these acoustic holograms.
My research focuses on deep learning for dense estimation problems such as optical flow. My broad interests include computer vision and machine learning.
My research in computer vision focuses on predicting age, gender and facial attractiveness of face images in the wild using deep learning. Other works include fine-grained object classification, such as car types, improving non-maximum suppression for object detection as well as image enhancement and single image super resolution.
I explore the connection between 3D reconstruction and semantic data, in particular, how 3D models can be improved if parts of it are segmented into classes or, vice-versa, how 3D world can be segmented based on shapes of objects. My work concentrates on jointly inferring dense 3D geometry and semantic labels from multiple images.
I will work in the area computer vision, machine learning and robotics, and specifically focus on SLAM methods (simultaneous localization and mapping) for medical robots.
I am currently working on magnetic helices as micro-/ nano-propellers, propulsion strategies in biological non-Newtonian environments at low Reynolds numbers, and related applications in biological fluidic systems.
Sebastian is an advocate of pragmatic causal modelling and aims at bringing statistical causal modelling from pen and paper to fruitful application. He pursues conceptual work on how our ability to causally reason about a system depends on the variables and transformations thereof being used as descriptors.
My research focuses on motion estimation from complex videos. I am especially interested in computationally efficient methods to compute the motion, and the interplay between the motion and the 3D structure of a scene.