DE / EN

SM 453/IS 722/IS 918 Bachelor and Master Seminar

Course Information
Lecturer Prof. Dr. Christian Becker
Type Seminar (SM 453 for Bachelor & IS 722 for Master & IS 918 for MMBR)
Credit Points 6 ECTS (MMM & MMBR), 5 ECTS (B.Sc. WI since HWS 2013), 4 ECTS (M.Sc. WI since HWS 2013 and B.Sc. WI before HWS 2013)
Prerequisites Basic knowledge in information technology
Course Language English
Form of assessment Conference style seminar (see details below)
Registration Please see information below!
Martin Breitbach, M.Sc.

Martin Breitbach, M.Sc.

Contact person for Seminar

For further information please contact Martin Breitbach.

  • Schedule

    Registration August 01 - September 20
    Confirmation until September 23, noon
    De-Registration possible until September 24, noon
    Kick-Off Meeting September 28
    First paper draft deadline November 13
    Review deadline November 20
    Camera ready (final) paper deadline November 27
    Conference (final presentations) December 04
  • Conference style seminar

    This seminar is organized in a scientific conference style. All accepted participants must write a scientific paper about the assigned topics and submit those papers until the first draft deadline. After that, the paper review phase starts and each paper will be assigned to at least two other participants who have to review the papers of two or three other authors. After the review phase, the reviews must be submitted to the supervisors, which distribute them to the paper authors. After that, the authors have time to improve their papers based on the feedback from the reviews, before they need to hand in their camera-ready (final) version of the paper. At the end of the semester, the „conference“ (final presentations) will take place.

    The grading is divided into different parts: The first part is the camera ready version of the seminar paper. This is the most important part and it is weighted with 50% of the overall grade. Second, the reviews for the other authors  are weighted with 20%. It is crucial to look at the work of others with a critical eye and to give constructive feedback. The last grading criterion is the presentation at the „conference“ and the participation during the discussions (30%). 

    If the assigned topic contains an implementation part, students do not need to participate in the review phase. For these topics, the grading is divided into 70% for the implementation and 30% for the presentation and participation during the discussions.

    Attendance at the kick-off session and the final presentation session is mandatory.

    All papers must use the IEEE manuscript template. We offer a customized version of the template here.

  • Registration

    Please apply via our online registration tool only (accessible inside the university network via VPN) . Registration is possible from August 01 until September 20.

    Requirements:

    • CV
    • Transcript of records

    We will not consider registrations via e-mail or incomplete registrations in the registration tool.

     

  • Topics

    01 - A Survey on Gesture Detection Algorithms based on Computer Vision

    Supervisor: Anton Wachner

    Unobtrusive touchless input methods based on gaze estimation are an emerging trend in human-computer interaction but most of the recommended systems still rely on dedicated hardware configurations. Computer vision-based techniques exist that only need the raw image data as input and allow gaze estimation based on the pure camera input frames. While input methods which rely on the point of gaze as triggering mechanism are prone to misinterpret “normal” eye movements as action, systems based on gaze gestures has been proven to work more robust. A key challenge is still to detect predefined movement patterns and gestures respectively in real-time.

    The objective of this seminar work is to categorize and compare techniques for image-based gesture detection based on a thorough literature review. Further, the applicability in the context of facial feature extraction should be discussed.

    02 - A Survey on Machine Learning Methods in the Field of Computer Vision

    Supervisor: Anton Wachner

    Human-computer interaction based on computer vision is an emerging trend in HCI research. Techniques for image processing and pattern detection can be applied to build touchless input systems that uses the raw image data from a camera video streams as input in order to detect gestures of a user standing in-front of the camera. However, prominent approaches are based on computationally intensive algorithms which affect the responsiveness of the overall input system and attractiveness of the input method for users. Light-weight machine leaning models has the potential to replace geometric computational models and provide the possibility to perform classification tasks for identifying gestures very efficiently.

    The objective of this seminar work is to identify and categorize machine learning methods for computer vision tasks based on a thorough literature review. Further, classification approaches should be compared and for each group of algorithms an overview should be given for which sort of problems they work best.

    03 - Unlimited Power: A Review of Code Offloading Approaches for Internet of Things Devices

    Supervisor: Martin Breitbach

    Modern applications in the areas of machine learning, augmented reality, or mobile gaming require running computationally intensive tasks on resource-constrained devices such as smartphones or smartwatches. One approach to overcome the limitations of the hardware is to offload tasks to more powerful remote devices such as laptops, desktop PCs, edge servers, or the cloud. Internet of Things (IoT) devices in smart homes, smart clothing, or Industry 4.0 often cannot apply code offloading as they are not able to run the software that performs code offloading itself or are only able to communicate via distinct communication channels such as Bluetooth Low Energy. However, recently, approaches emerge which bridge the gap between the IoT and the cloud and enable IoT devices to perform code offloading.

    The goal of this seminar thesis is to identify and evaluate code offloading approaches for IoT devices with a thorough literature review. How do these approaches connect the IoT to edge and cloud? What are the differences between approaches? How does an ideal offloading approach for the IoT look like? 

    04 - Offloading for Future: An Overview of Energy-Aware Code Offloading Approaches

    Supervisor: Martin Breitbach

    Apart from improving the computational performance, code offloading can help reducing the energy consumption of a device. Instead of executing tasks locally, resource-constrained devices may offload tasks to more energy-efficient resource providers such as other desktop PCs, edge servers, or the cloud. Many approaches aim to minimize the energy consumption of one battery-powered device such as a smartphone with code offloading. However, they neglect the perspective of the global energy consumption of the entire distributed computing system. This “global” energy consumption can be reduced by allocating tasks carefully to energy-efficient devices.

    The goal of this seminar thesis is to present, compare, and evaluate code offloading approaches that minimize the total energy consumption of a computing system. How do the approaches differ and how do they minimize energy consumption? Do they consider performance as an additional goal? What are incentives for people to participate in such systems?

    05 - A deep dive into adaptation rules for adaptive interfaces.

    Supervisor: Melanie Heck

    Adaptive interfaces define user models based on which the interface dynamically adapts to the current context. In order for the system to behave in such a way that it supports the user, an adaptation logic has to be defined. This adaptation logic typically specifies what (i.e., adaptation constituents), when (i.e., adaptation determinants), why (i.e., adaptation goals), and how (i.e., adaptation rules) to adapt.

    The objective of this seminar paper is to devise a taxonomy on adaptation rules for adaptive interfaces. Based on this taxonomy, prevalent adaptive interface systems should be classified, thus providing a comprehensive overview of the existing literature.

    06 - Coping with cognitive overload: A survey on use cases for adaptive interfaces.

    Supervisor: Melanie Heck

    Adaptive interfaces monitor the user and the environment in order to dynamically adapt the interface to the current context. These kinds of interfaces help users deal with information overload, e.g., in online shops (recommender systems), web browsers (personalized search engines), or e-learning environments.

    The objective of this seminar paper is to identify application areas for adaptive interfaces and provide an overview of the systems that have been developed for each of the identified areas.

    07 - From gaze to personality: A survey on gaze data processing approaches for user profile creation.

    Supervisor: Melanie Heck

    Using eye tracking to evaluate attentional parameters has become an established practice in market research to investigate consumer attention to products and their features. As the technology becomes more mature, companies like Apple and Google have worked on integrating eye tracking functionalities into their devices. This opens up new opportunities to unobtrusively track the user’s attention and learn something about their interests or even their personality. Since many factors influence the users’ gaze as they naturally interact with their devices, the gaze cannot be used directly to create a user profile. Filtering mechanisms and machine learning algorithms are therefore applied to infer user characteristics.

    The objective of this seminar paper is thus to conduct a survey on approaches that allow to establish user profiles based on gaze data in a natural setting.

    08 - A Taxonomy for Cyber-Physical Systems

    Supervisor: Melanie Brinkschulte

    As computing devices became more and more mobile, the research area of cyber-physical systems (CPS) emerged out of the area of Pervasive Computing.  Nowadays CPS are ubiquitous and are used in a wide range of different application fields. For example, they can be found in smart-home environments, remote patient monitoring or in the automotive sector. Due to their many and varied applications, CPS are a highly relevant topic.

    The objective of this seminar paper is to create a taxonomy to provide a conceptual framework for future analyses of CPS in different application areas. Furthermore, existing CPS should be classified based on this taxonomy to provide a comprehensive overview of existing literature.

    09 - Cyber-Physical System Simulations

    Supervisor: Melanie Brinkschulte

    Cyber-physical systems (CPS) have lately gained immense popularity as being a key part of the realization of, e.g., Smart Cities, Industry 4.0 and autonomous driving. As the name implies, the focus in such systems is on the combination of cyber (computation and communication) and physical (sensors and actuators) components which interact in a feedback loop. Furthermore, in most application fields CPS cooperate with each other by exchanging information (processed sensor data, derived data, etc.) over a network to fulfill their tasks and achieve their goals. Therefore, the systems are composed of multiple networked components. This leads to a highly complex and distributed control system. As a result, the development, testing and maintenance is rather complex.

    The objective of this seminar paper is to identify the challenges in CPS design, testing, maintenance and simulations. Furthermore, an overview of existing simulation approaches, including their aims and used tools, should be provided.

    10 - Rules, Goals and Utilities for Specifying Self-Adaptive Systems

    Supervisor: Martin Pfannemüller

    Self-adaptive Systems modify themselves at run-time to maintain their performance after changes in the system resources or the environment. This is called adaptation. For specifying the adaptation behavior simple rules can be used. On the other end of the spectrum, goals and utilities can be used for defining a system’s behavior. Also, hybrid approaches combining, e.g., rules with utilities can be used.

    The goal of this seminar paper is to review self-adaptive systems approaches using rules or higher-level goals and utilities for specifying the behavior. This work should describe the three types of decision criteria and compare the found approaches using a taxonomy in terms of their capabilities.

    11 - Simulating Renewable Energy Communities

    Supervisor: Sonja Klingert

    In order to spur the energy turnaround targets set by the Paris Agreement in 2015, the EU has created the instrument of Renewable Energy Communities (RECs).  These are institutions that are allowed to produce, consume, store and sell renewable energy with the primary purpose providing environmental, economic or social community benefits for its shareholders. In order for RECs to become a success story they need to be financially, socially and energetically sound.

    The goal of this seminar paper is to review existing simulation frameworks or tools that aim at representing at least two of the mentioned dimensions.

    12 - Energy Communities in Europe – Status and Evaluation

    Supervisor: Sonja Klingert

    In order to spur the energy turnaround targets set by the Paris Agreement in 2015, the EU has created the instrument of Renewable Energy Communities (RECs).  These are institutions that are allowed to produce, consume, store and sell renewable energy with the primary purpose providing environmental, economic or social community benefits for its shareholders. In order for RECs to become a success story they need to be financially, socially and energetically sound.

    The goal of this seminar paper is create an overview (incl. a short evaluation) about existing energy communities in Europe.