Studienund Masterarbeiten Supercomputing Systems AG
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Studien- und Masterarbeiten Supercomputing Systems AG Stand Oktober 2019 https://www.scs.ch/karriere/studien-und-masterarbeiten-bei-scs/ Vision trifft Realität. Supercomputing Systems AG Phone +41 43 456 16 00 Technopark 1 Fax +41 43 456 16 10 8005 Zürich www.scs.ch
Towards User-independent 2D/3D Object Classification of Complex Life Science Images We develop novel machine vision & mathematical morphology algorithms to analyze complex multi- modal Life Science images. The goal of this work is the development of 2D/3D image segmentation & classification algorithms for (semi-)automated analysis of complex cellular structures in plants & liver images from Electron Micrographs (EM). Art der Arbeit: 30% Theory, 70% Software Engineering Voraussetzungen: Solid knowledge of Matlab and C++ Good knowledge of Image Processing Aufwand: MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 2 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Automating DDH Diagnosis using Machine/Deep Learning Techniques Between 2 and 3 percent of all infants are diagnosed with developmental dysplasia of the hip (DDH), and it is believed that approximately 30% of all hip replacement surgeries on patients below the age of 60 are owed to DDH. While previous scientific publications as well as research at SCS has shown that deep learning-based methods can already outperform human experts in measuring angles of the hip bones, there is no method yet to reliably detect automatically whether an US image has been obtained using correct orientation of the US probe. Art der Arbeit: 60% Research, 20% Benchmarking, 20% SW Development Voraussetzungen: Machine Learning & Deep Learning Basics, Python, C/C++ Aufwand: MA, 1 Person Ansprechperson : fabian.schenkel@scs.ch 3 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Transfer Learning for Image Segmentation using Convolutional Neural Networks Today’s Machine Learning (ML) success is often limited by a lack of labelled ground truth data to train the models. This is especially true for applications in medical imaging. In medical optical coherence tomography (OCT) of the eye, only a limited set of GT- labeled images exists. Transfer learning (TL) is a state-of-art ML-technique that can be useful to overcome this problem for similar yet distinct tasks. Art der Arbeit: 30% Theory, 40% Implementation, 30% Evaluation Voraussetzungen: High-level programming language, ML/DL knowledge helpful Aufwand: MA, 1 Person Ansprechperson : fabian.schenkel@scs.ch 4 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
MLOps: Bring Your Model to Production The field of ML is growing up. In the past, questions like "which AI algorithm should I choose" and "what is the best way to train it" were the main focus of AI projects. Today, these topics are well explored. Now, new questions get into focus: how can we bring our model actually to production in a reliable and reproducible way? How can we integrate data acquisition, training, and monitoring of our model into an automated system? Art der Arbeit: 60% Implementation, 20% Theory, 20% Testing/Benchmark Voraussetzungen: Machine Learning Basics, DevOps Basics Aufwand: MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 5 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Deep Learning in the Wild Climate change and human exploitation of our planet has a significant influence on the habitat and existence of wild animals. The resulting biodiversity loss threatens ecosystems and the human development that depends on them. Protecting these habitats is based on delivering evidence by collecting data. This is usually labour intensive, since it depends on field work done by biologist and volunteers. This master thesis tries to make a contribution to scale up this important process by using acoustic detection of animals using deep learning on embedded systems. Art der Arbeit: 50% Deep Learning, 30% Embedded Dev, 20% Benchmark Voraussetzungen: ML & DL Basics, Signal Processing, Python Aufwand: MA, 1 Person Ansprechperson : fabian.schenkel@scs.ch 6 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Deep Learning for Estimating the Rail-Wheel Forces of an ICN based on Track Geometry Tilting trains can travel at higher speeds without degrading the passenger comfort. However, these higher speeds lead to larger rail-wheel interaction forces. This larger forces have to be monitored in order to maintain general safety (e.g. to prevent derailment). This thesis tries to make a contribution to a cost efficient monitoring system. The goal is to learn the train characteristics with a deep learning network so that it can process track geometry data and predict the resulting forces. Art der Arbeit: 30% DL Theory, 15% Theory Rail/Wheel Interaction, 40% Implementation, 15% Benchmark Voraussetzungen: ML & DL Basics, Signal Processing, Python Aufwand: BA/MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 7 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Automatisierte Anamnese von Heizungen und Gebäudeenergieanlagen In der Schweiz besteht ein grosses Energieeffizienzpotential in Gebäuden. Bestehende Heizungsanlagen sind heute oft nicht ideal eingestellt, weisen Mängel auf oder sind gar dysfunktional. Zur Detektion dieser Mängel soll ein Prototyp eines automatisierten Anamnese-Systems aufgebaut werden, welches aufgrund von sensorischen Messgrössen (primär Temperaturfühler) die Anlage identifiziert, auf Fehler analysiert und eine Liste von Handlungsempfehlungen generiert, um das System zu verbessern. Art der Arbeit: 40% Theorie, 60% Umsetzung Voraussetzungen: Signalverarbeitung, Programmierung Aufwand: MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 8 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Detektion von Kabelfehlern in der Aussenanlage Eine Herausforderung beim Betrieb von Bahnstrecken ist die Überwachung von Elementen (z.B. Weichen) und deren Verkabelung. Ein weit verbreitetes Verfahren ist die Überwachung der Adern durch einen Ruhestrom. Speziell bei den heute immer noch weit verbreiteten Lösungen auf Relaisbasis ist die Überwachung des Stromes sehr eingeschränkt. In der vorliegenden Arbeit soll ein Verfahren konzipiert, aufgebaut und bewertet werden, das es erlaubt, rückwirkungsfrei einen Rückschluss auf die bestehende Verschaltungen der vorliegenden Aussenanlage zu geben, und damit den Betrieb und Instandhaltung bei der Fehlersuche unterstützen. Art der Arbeit: 30% Theorie, 40% Algorithmus, 30% Erprobung Voraussetzungen: Signalverarbeitung Aufwand: MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 9 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Self-Calibration for Embedded Stereo Vision System Stereo Vision allows accurately measuring scenes in 3D – provided that the cameras are well calibrated. Harsh industrial or automotive contexts affect opto-mechanical properties of the stereo-rig and its measurement accuracy. The goal of this thesis is to explore, test and analyze different methods and algorithms for continuous stereo- vision calibration. The solution has to determine the extrinsic and intrinsic camera parameters without any markers or reference objects. Art der Arbeit: 40% Theory, 40% Implementation, 20% Benchmarking Voraussetzungen: Computer Vision, OpenCV Aufwand: MA, 1 Person Ansprechperson : fabian.schenkel@scs.ch 10 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Virtual Reality as 3D Ground Truth Generator for AI, Machine Learning and Deep Learning Training effective Artificial Intelligence (AI) algorithms today often requires large amounts of ground truth data. Typically, this is a laborious, costly and time- consuming process. These problems can be overcome by combining AI with Virtual Reality (VR): VR creates artificial environments that often resemble our real world. This project aims at exploring the potential of VR as 3D ground truth generator for state-of-the-art Machine Learning (ML) and Deep Learning (DL) algorithms. Art der Arbeit: 50% Research, 20% SW Development, 30% Benchmarking Voraussetzungen: Machine Learning & Deep Learning Basics, Python, Matlab Aufwand: MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 11 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Generating 3D Indoor Maps Autonomously Through Reinforcement Learning In September 2017, Unity Technologies released the first open beta of the Unity Machine Learning Agents Toolkit. With this toolkit, it is possible to train agents to solve a specific task in a simulated environment. But how complex can a task in such a simulated environment be, which we want to solve? To explore this issue, we want to train an agent using reinforcement learning that can be used to autonomously move an unit through a room. At the same time, the agent should be able to scan the room and generate a 3D map of it. Art der Arbeit: 30% Theorie, 70% Umsetzung Voraussetzungen: Machine Learning, CV Basics, Unity Basics, C#, Python Aufwand: MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 12 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Decentralized Ledger eVoting System eVoting is an unsolved problem, mainly because of security concerns connected to centralized IT systems. Blockchain technology enables new ways to solve IT problems in a transparent, tamper-proof and decentralized way. This master thesis aims at implementing a voting solution that scales to Swiss national votes. The technologies to be used could include zero knowledge proofs (zk-SNARKS), homomorphic encryption and smart contracts on a public blockchain (i.e. ethereum). Art der Arbeit: 60% Theory, 40% Implementation Voraussetzungen: Cryptography, Blockchain Aufwand: MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 13 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Building a new Ecological and Private Cryptocurrency encointer proposes a new blockchain-based cryptocurrency with an ecological consensus mechanism using trusted execution environments and an egalitarian money supply policy, where money issuance is done by individuals attending randomized pseudonym key signing events. encointer also features scalable private transactions and trustless off-chain smart contracts. This thesis shall build an encointer testnet based on Hyperledger Sawtooth. Art der Arbeit: 30% Theorie, 70% Umsetzung Voraussetzungen: Blockchain basics, Go Aufwand: MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 14 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Software-Architecture for SDR (Software Defined Radio) Alle aktuellen SDR Softwarepakete haben Einschränkungen. Es gibt GNU Radio, welches kompliziert und schwierig zu bedienen ist, und es gibt verschiedene SDR GUIs, welche jedoch nicht einfach erweiterbar sind. Das Ziel ist eine einfach zu bedienende Software zu erstellen, welche es ermöglicht, Radiosignale zu analysieren und Algorithmen auszuprobieren. Fertige Abläufe sollen dann auf Knopfdruck in eine C-Datei exportiert werden können, um eine Integration in Embedded Systeme so leicht wie möglich zu gestalten. Art der Arbeit: 10% Theorie, 90% Software Engineering Voraussetzungen: C/C++ oder C#, GUI design (MVVM, WPF, QT oder Web) Aufwand: SA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 15 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
AI-based Golf-Coach with Automated Swing Analysis Der Golfschwung gilt als einer der komplexesten Bewegungsabläufe aller Sportarten. Für den Amateurgolfer besteht die Schwierigkeit darin, bei einem Fehlschlag den Fehler in seiner Bewegung zu erkennen und zu korrigieren. Das Ziel dieser Arbeit ist die automatisierte Analyse des Bewegungsablaufs des Golfspielers anhand von Filmaufnahmen, gezieltes Erkennen fehlerhafter Muster, sowie das Ausgeben von Korrekturvorschlägen. Art der Arbeit: 20% Theorie, 70% Umsetzung, 10% Benchmarking Voraussetzungen: Interesse an Computer Vision und SW-Entwicklung Aufwand: SA/BA/MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 16 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Flexible FPGA-based Test Equipment to Model and Characterize the Real-Time Behavior of an Ethernet Ring In this work, flexible test equipment for Ethernet network hardware and firmware shall be evaluate and commissioned. The goal is to model and characterize an Ethernet network based on BroadR-Reach with ring topology and HSR protocol for its real-time behavior. The test equipment could be based on devices available at http://netfpga.org. Extensions may be developed to further improve functionality. Art der Arbeit: 40% Theorie, 60% Implementation Voraussetzungen: Ethernet Networks, FPGA Implementation, SW, Testing Aufwand: SA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 17 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Jogging with Acoustic Feedback based on Body Rhythms Jogging has become very popular in recent years, with millions of people across the world integrating it into their regular exercise regime. Many, however, do so with poor movement coordination, particularly in terms of the synchronization of body rhythms such as cadence, breathing and heartbeats. In this Master’s thesis, we aim to develop a digital application that will support runners in finding their natural jogging ‘groove’ by providing them with acoustic, real-time feedback on their individual style and technique. Art der Arbeit: 20% Theory, 60% Design/Implementation, 20% App Dev Voraussetzungen: Programming basics, Enthusiasm Aufwand: MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 18 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Privacy Preserving OAuth Service with TEEs If we use Google, Facebook, or SwissID logins for third party sites, these login providers get to know when we log into which service. Moreover, the user has only little control over which data from her/his Google or Facebook accounts are shared with the service she/he's logging into. The goal of this thesis is to build a blind Oauth service using a Trusted Execution Environment, which can be easily integrated into third party sites just like the above mentioned services. Art der Arbeit: 30% Theory, 70% Implementation Voraussetzungen: Rust Aufwand: MA, 1 Person Ansprechperson : fabian.schenkel@scs.ch 19 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Enhanced Body Control through Body Tracking and Data Visualization Currently, especially athletes and tech geeks are using tools that allow to measure and track body functions like pulse, blood pressure, or respiration. However, the trend towards measurement of bodily functions is increasing steadily, while the possibilities of data acquisition are constantly being extended. The goal of this work is to develop a new tool that allows a human to learn to control non-visible body functions through visual feedback provided by the tool. If the project is successful, the goal is to show the final work in an art exhibition. Art der Arbeit: 40% Theory, 60% Implementation Voraussetzungen: Data visualization Aufwand: SA/BA/MA, 1-2 Personen Ansprechperson : fabian.schenkel@scs.ch 20 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Developing a Payed Web Service that Guarantees no Data Collection What happens with our data when you log-in to an online service, e. g. over a smartphone app? Is your complete usage profile forwarded to the company? The goal of this thesis is to evaluate an approach that obfuscates the actual usage of a service, while maintaining the possibility to restrict access to paying customers. The idea is to run the login process on a Trusted Execution Environment. What happens inside the TEE cannot be seen by the company running the service, not even by an admin. Art der Arbeit: 30% Theory, 70% Implementation Voraussetzungen: Interest in privacy, Openness to work with new technologies Aufwand: MA, 1 Person Ansprechperson : fabian.schenkel@scs.ch 21 Zürich 07.10.2019 © by Supercomputing Systems AG CONFIDENTIAL
Supercomputing Systems AG info@scs.ch +41 43 456 16 00 Vision meets reality. Supercomputing Systems AG Phone +41 43 456 16 00 Technopark 1 Fax +41 43 456 16 10 8005 Zürich www.scs.ch
You can also read