Max Planck ETH Center for Learning Systems| Acronym | MPG ETH CLS | | Homepage | http://learning-systems.org/ | | Country | [nothing] | | ZIP, City | | | Address | | | Phone | | | Type | Alliance | | Current organization | Max Planck ETH Center for Learning Systems | | Members | |
Open OpportunitiesIn continual learning, deep learning models incrementally learn more classes or tasks over time. Doing so, they should not forget previously learned knowledge. This is a hard and active research problem. Making it even harder, we want the models to also estimate correct uncertainty. E.g., they should be highly uncertain about a new object type, but not uncertain about an object that they just learned correctly.
[1] Parisi et al., Lifelong learning with neural networks http://dx.doi.org/10.1016/j.neunet.2019.01.012
[2] Gawlikowski et al., Uncertainty in Deep Neural Networks http://arxiv.org/abs/2107.03342 - Engineering and Technology, Information, Computing and Communication Sciences
- Master Thesis, Semester Project
| Rotation optimization occurs in camera calibration, kinematics, animation, and attitude setting of spacecraft. These optimization problems are generically nonconvex and difficult to solve, but some problem variations offer exact solutions in the low-noise and outlier-free setting. This project will investigate Harmonic Hierarchies of polynomials as a mechanism to perform global optimization of outlier-robust rotation alignment tasks. Harmonic Hierarchies utilize the properties of the rotation space and binary hypercube to produce a sequence of linear programs in increasing size, yielding a sequence of convergent upper-bounds and lower-bounds to the true alignment error. Extensions to this project can include multi-camera alignment (group averaging) and problems in inverse kinematics.
This project will be performed in collaboration with Lucas Slot (ETH Zurich: D-MATH) and Mauricio Velasco (Universidad Catolica del Uruguay). - Engineering and Technology, Mathematics
- ETH Zurich (ETHZ), Master Thesis, Semester Project
| In this project, co-supervised by Swissgrid, we will look into the problem of designing incentives for the procurement of voltage support services in the grid. The design of an efficient mechanism requires the use of tools from control theory, game theory, and optimization. - Electrical Engineering
- Collaboration, Energy (IfA), Master Thesis
| The advertised project focuses on the implementation of
meta-learning based control approaches on a ball-on-a-plate system, which requires redesigning
the system’s electronics. Hence, this project combines electronics, control theory, image recognition
and machine learning. - Artificial Intelligence and Signal and Image Processing, Electrical and Electronic Engineering
- ETH Zurich (ETHZ), Master Thesis, Semester Project
| Modern robots collect data from various sensors. When these sensors operate independently, time-synchronization through rectification of their individual clocks and correction for temporal drift is required.
In our previous work, we developed an initial version of a synchronization pipeline in Python, designed for offline data synchronization. Our current pipeline already effectively synchronizes sensors that include a common external synchronization signal. Despite already working well, our current pipeline still requires some expertise to configure the data sources. To make the pipeline widely usable, we now need to make it function seamlessly even without expert knowledge and access to external synchronization signals. This enhancement should also extend to scenarios involving continuous online data as well.
Furthermore, we want to prove the correctness of the synchronization and showcase the performance based on synthetic data.
In essence, your thesis will comprise the following key objectives:
1. Understand the challenges involved in data synchronization.
2. Familiarize yourself with the existing synchronization pipeline.
3. Innovate strategies for achieving data synchronization without relying on external synchronization signals.
4. Enhance the user interface by creating an intuitive guide for using the pipeline effectively.
5. Extend the functionality to accommodate online data streams.
6. Assess the pipeline's correctness and performance using synthetic biosignals, as well as pre-recorded biosignals from the SMS-Lab and Tohoku University.
Throughout this project, you will receive guidance from me, a 4th year PhD candidate at the Sensory-Motor Systems Lab at ETH Zurich, and researchers at Tohoku University in Sendai, Japan. As I will be in Japan from October, we will conduct the weekly meetings over Zoom.
Furthermore, in case of interest, you have the exciting opportunity to visit us in Japan. This opportunity can be pursued either through personal funding or by applying for respective scholarships, such as the Heyning-Roelli Foundation, SEMP, Spickenreuther Foundation, and others. I have received scholarships in the past and I am happy to provide guidance and support throughout the application process. - Biosensor Technologies, Data Storage Representations, Data Structures, Digital Systems, Information Storage, Retrieval and Management, Pattern Recognition, Signal Processing, Simulation and Modelling, Software Engineering
- ETH Zurich (ETHZ), Internship, Master Thesis, Semester Project
| In this project, we focus on continuous and quantitative monitoring of activities of daily living (ADL) in SCI individuals with the goal of identifying cardiovascular events and PI-related risk behaviors.
ADLs specific to SCI patients and their lifestyles shall be discussed and narrowed down in the scope of this work, therefore an autonomous camera-based system is proposed to classify ADLs.
The Current work builds on a previous project where a SlowFast network [1] was trained to identify SCI-specific classes and we aim to further improve the classification and temporal resolution for transferring to wearables' time-series data. - Computer Vision, Health Information Systems (incl. Surveillance), Intelligent Robotics, Knowledge Representation and Machine Learning, Neural Networks, Genetic Alogrithms and Fuzzy Logic, Pattern Recognition
- Bachelor Thesis, ETH Zurich (ETHZ), Internship, Master Thesis, MD
| We simulate the tracking problem with a Franka Emika Panda robot
in PyBullet. We want the robot to precisely track a cubic object moving on the conveyor belt. In this
project, we divide the tracking of a moving object into two phases. The first phase is the reaching phase,
where we estimate a physics model of the robot to design a controller. The second phase is the tracking
phase where we want to train a reinforcement learning policy (e.g., deep deterministic policy gradient
(DDPG)) for the robot given the perturbed target location. - Control Engineering, Electrical and Electronic Engineering, Intelligent Robotics, Robotics and Mechatronics, Systems Theory and Control, Systems Theory and Control
- ETH Zurich (ETHZ), Semester Project
| Robots need to manipulate a wide range of unknown objects, from transparent to shiny surfaces. The goal of this project is to investigate learning techniques to bridge the visual domain gap between high-fidelity rendered scenes and real-world images for scene understanding.
- Computer Vision, Intelligent Robotics, Knowledge Representation and Machine Learning
- Master Thesis
| One of the new disciplines at the upcoming CYBATHLON is the vision assistance race. Visual impaired people are severely limited in their autonomy of completing many daily tasks. Available vision aids are limited to specific domains, such as reading text out loud, but fail to generalize. Smart vision assistive technologies could provide more intuitive, comprehensive and reliable support in daily tasks.
The CYBATHLON challenges contain a variety of daily situations, such as shopping, finding a free seat or ringing the correct doorbell. The goal is to develop an assistive device capable of fulfilling all challenges.
- Computer Vision, Image Processing, Intelligent Robotics, Mechanical Engineering, Neural Networks, Genetic Alogrithms and Fuzzy Logic
- Semester Project
| Diffusion models have a huge potential in motion planning and navigation. In this project, we focus on generating spline-based trajectories using diffusion models able to make ANYmal navigate in extremely challenging dynamic environments - Intelligent Robotics
- Master Thesis, Semester Project
|
|
