Maschinelles Lernen im Energiesektor

In der Projektgruppe "ML@Energy" soll eine Anwendung entwickelt werden, die Analysen von Energieverbrauchsdaten visuell darstellt. Die Analysen basieren dabei auf verschiedenen Machine Learning Verfahren, die innerhalb der Projektgruppe angewendet werden.
Außerdem werden agile Methoden des Software Engineerings angewendet.

Raising Awareness on the Privacy Impact of MR Headsets

Mixed Reality (MR) headsets offer great opportunities for a wide range of applications for leisure, work, education, and marketing, among others. With MR, users can be immersed in a virtual world (Virtual Reality) or augment their view of the real world with virtual content (Augmented Reality). However, MR headsets utilize an array of sensors (including many cameras) and bring them closer to the human body, enabling sensitive data to be captured, processed, and shared with third parties. State-of-the-art headsets already provide access to behavioral data (e.g., hand and body motion, eye gaze), physiological data (e.g., electroencephalography, heart rate), contextual data (e.g., size of tracking space, bystanders), and device specifications. Such data allow one to infer information about user demographics (e.g., age, gender, handedness), health and well-being (e.g., reaction times, fitness level), and impairments (e.g., visual or motor impairment). Most users are not aware of these privacy implications that the utilization of MR devices entails.

The goal of this project is to build PlatforMR – a platform capable of receiving the data that is recorded by the various AR and VR applications during the deployment of different MR headsets. Therefore, interfaces accessing the data recordings of the headsets, intercepting the data flow between applications, and transmitting the captured data to the platform also have to be implemented for different devices. The intention of PlatforMR is to present the recorded data to users after interacting with the headsets, inform them about the potential information that can be derived from it, and evaluate the user’s perception of these privacy implications of MR devices.

A Platform for Regional Mobility Analysis

Personalized motorized transport is an important source of carbon dioxide emissions. With initiatives such as the 49€ ticket, the federal government is actively incentivizing citizens to switch to public transportation to reduce the carbon dioxide footprint of their everyday mobility. However, despite this cost reduction for public transportation realized by the 49€ ticket, many citizens are reluctant to change their habits. Besides from reduced convenience, many persons state that the mobility offers available in their region do not match their mobility needs or do not exhibit the necessary reliability as short delays encountered during a trip can stack up quickly, e.g., due to missed connections, etc.

The goal of this project group is to develop a prototype platform to analyze the structure of the public transportation offers available in a region to systematically identify gaps and potentials for improvement. For the project group, the focus of the analysis lies on the statistical characteristics of the mobility offers that are available in the Ruhr area during a certain time frame. Towards this end, the platform will collect real-time data for different mobility offers and derive statistical models for their availability. Using these models, the platform will compute a reachability matrix that reflects the expected trip durations during different time periods which can then be combined with other data to identify optimizations.

From a technical perspective, the project group will encompass the development of statistical models for the reliability and availability of mobility offers, the collection and processing of large volumes of data, the development and customization transit routing algorithms and the visualization of spatial data. As basis for the implementation, we are planning on reusing existing software libraries that are written in Java.

From a theoretical perspective, the project group will cover concepts related to the processing of spatial data, such as routing, visualization, etc. In addition, the participants will prepare individual seminar talks and papers on selected research topics related to mobility.

Härtung von Analysesandboxen gegenüber Schadsoftware für Linux

Täglich müssen Anti-Viren-Hersteller tausende neue Schadprogramme untersuchen. Eine manuelle Analyse aller Programme ist daher nicht möglich. Deswegen gibt es verschiedene Ansätze Analyse von potentieller Schadsoftware zu automatisieren. Dies geschieht im Regelfall durch die Nutzung von sogenannten Sandboxes oder virtuellen Maschinen. In diese wird eine potentielle Malware in einer isolierten Umgebung ausgeführt, um schadhaftes oder unerwünschtes Verhalten zu erkennen. Um eine Analyse zu erschweren, versuchen daher Malwareentwickler, eine Ausführung in einer Sandbox zu entdecken. Bei erfolgreicher Erkennung wird kein verdächtiges Programmverhalten offenbart. Die PG HoLMAS wird Malware für Linux Betriebssysteme untersuchen, mit besonderem Fokus auf Verschleierungstechniken, um Malware-Analyse zu erschweren. Hierzu werden verschiedene Schadprogramme als auch spezielle Testsoftware untersucht werden. Eine bereits bestehendes Framework zur Analyse von Linux Malware soll so angepasst werden, dass es einer Malware schwer fällt die Analyseumgebung zu erkennen. Dabei werden die Studierenden sowohl bestehende als auch neue Methoden entwickeln, um eine VM zu erkennen bzw. eine Erkennung zu verhindern

Learning Point Cloud infilling for Humans

In this project, we want to record humans as 3-dimensional point clouds using multiple depth cameras. However, as only a limited number of cameras is available, we will always have to deal with occlusion, meaning sections of the human which can not be captured by the depth-cameras. For example, this happens when an arm covers part of the torso.

To address this issue, we want to learn how to generate point cloud patches which fill in occluded sections of the recordings, allowing to derive watertight surfaces. Predicting these patches is done by combined estimation of the geometry and the texture of the missing sections in the recording.

To demonstrate the system, a capturing demo should be implemented using multiple Intel RealSense Depth Cameras (Hardware-Setup will be provided). The recordings of multiple cameras need to be combined into a single Point Cloud. First, the human needs to be segmented from the background. In the main step, the Point Cloud will be completed in a learned infilling module.

• Ausführliche Informationen zu den einzelnen Projektgruppen gibt es unter Moodle.
• Nicht-UDE-Studierende ohne UDE-Unikennung müssen sich für Moodle erst registrieren. Den Einschreibeschlüssel fragen Sie bitte bei den Ansprechpartnerinnen und Ansprechpartnern nach.