Zum Inhalt springen
  • de
  • en
Menü
Technische Universität München
Technische Universität München

Advanced Seminar Embedded Systems and Internet of Things

We will announce the available topics and the application process on the 20th of September 2024. You will then be able to apply for a seminar topic between the 20th of September until the 4th of October (23:59 pm). 
It is mandatory to attend all the lectures of our Advanced Seminar in presence to complete the course successfully. Virtual Attendance is not possible. 

Application Process

 

Due to the high interest in our seminar topics we use an application process to assign the topics.

If you are interested in one of the topics below, please send your application together with your CV and your transcript of records to seminar.esi.ei@tum.de. Express your interest and explain why you want to have that specific topic and why you think that you are most suitable for the topic. This allows us to choose the most suitable candidate for the desired topic to maximize the seminar's learning outcome and to avoid dropouts.

Additionally, you can indicate a second topic that you would like to take, such that we can still find a topic for you if your primary choice is not available.

Deadline: We encourage you to apply until the 04.10.2024. Afterwards we will assign the topics and notify all applicants. After this date, we will answer to requests within 3 days (until 9th of October), assuming that there is enough motivation for the given topic. Once you are given the topic, we will ask for your confirmation. You must confirm your participation until the 11th of October.

Note: We do not assign topics on a first-come-first-served basis. Even though we appreciate your interest if you have asked or applied early for a topic we can not guarantee that you get a seat. Generally we have 3-4 applicants per topic. Please think carefully if you are able to do the work required as we have to reject other students. Generally, email clients remember the people you have communicated with. You will be registered to the seminar course by the Advanced Seminar Manager after the Kick-Off meeting on 16th of October. 

 

Kick-off meeting

This semester the seminar will be conducted in physical mode. This means that you must join the physical classes and presentation which you will find on the Moodle page. Additionally, you can schedule weekly meetings with your supervisor via Zoom or on campus. Lecture materials and videos will be available on Moodle.

The kick-off meeting will be on the 16th of October at 9:45 on Campus. We ask all successfully selected participants to be present in the kick-off meeting. Please notify us in case you can not make it to the meeting, otherwise we will assume that you are no longer interested and give your place to another applicant.

 

Topics

This semester we offer the following 7 topics for the advanced seminar "Embedded Systems and Internet of Things":

1. 3D space localization
2. Mathematical Schedulability Analysis of Multi Cyclic Shaper on Combined Shaper
3. AoI Based Method NFV-enabled network for scheduling wired-wireless communication
4. Open Source LLMs for local inference
5. Goal Management in Multi-Objective Optimization (MOO) with Pareto Reflections (Paref)
6. Context-Aware Adaptive Optimization: Leveraging Large Language Models in Multi-Objective Production Systems

7. Evaluation of Privacy Preserving Technologies in Healthcare: A Review of Related Work and Practical Applications

You will find the description of the topics below. Furthermore, we gave you a few references for each topic as a starting point for your research. Your task for each topic will be to read and analyze related literature, get an overview of the current state-of-the-art and summarize your findings in a paper-style report. Afterwards you will present your findings in a "mini-conference" in front of your fellow students.

During the seminar you will also learn through the lecture how to conduct the research, how to write a scientific paper and how to present your work.

1. 3D space localization

Description:

In the field of IoT and robotics, the ability of devices to accurately identify and communicate their positions relative to one another in a 3D physical space is important in autonomous systems, smart environments and manufacturing. The integration of localization methods plays a pivotal role in enabling effective coordination and interaction between IoT devices.

In this seminar topic, the student should investigate different localization approaches of IoT and robotics, comparing them regarding the necessary information and equipment, accuracy, application scenarios, and other relevant measures.

References:

Supervisor: Roman Binkert

Assigned

2. Mathematical Schedulability Analysis of Multi Cyclic Shaper on Combined Shaper

In this seminar topic, the student will study and propose the configuration optimization of cyclic shapers and multi‐shaper TSN networks, targeting the TAS + Multi‐CQF combinations such as TAS+MCQF, TAS+MCQF+CBS, TAS+MCQF+ATS, TAS+CQF+CBS, etc.

The seminar topic will involve the following tasks:

  1. Read and understand the shaping mechanisms of TSN. Understand the mathematical analysis.

  2. Extend the paper to the all TSN shaper combination and calculate the worst-case delay mathematically.

References:

Supervisor: Rubi Debnath

Assigned

3. AoI Based Method NFV-enabled network for scheduling wired-wireless communication

Description: In this seminar topic, the student will go through the AoI based method for joint resource allocation in wired-wireless communication.

The seminar topic will involve the following tasks:

  1. Read and understand the AoI based mechanism in TSN and 5G, focusing on the benefits of the AoI and the potential use case.

  2. Extend the paper to utilize the AoI for TSN cyclic shaper and 5G scheduling using mathematical validation and proofs.

References:

Supervisor: Rubi Debnath

Assigned

4. Open Source LLMs for local inference

Description:

In the era of Industry 4.0, artificial intelligence and machine learning are integral to advancing manufacturing processes. Open-source Large Language Models (LLMs) that can be run locally and fine-tuned present manufacturers with unique opportunities to enhance efficiency, optimize operations, and maintain strict control over proprietary data.

In this seminar topic, the student should investigate different open source LLMs and compare them regarding performance and their ability to fine tune, especially under the view of restricted hardware for local fine tuning and inference.

References:

Supervisor: Roman Binkert

Assigned

5. Goal Management in Multi-Objective Optimization (MOO) with Pareto Reflections (Paref)

Description: In modern dynamic systems, such as production environments, managing multiple conflicting objectives like environmental impact, throughput, and system reconfiguration is a complex task. Efficient goal management in MOO requires the ability to prioritize certain objectives while maintaining flexibility to adapt to changes in external conditions. This process often involves translating high-level goals into specific, quantifiable performance metrics.
Pareto Reflections (Paref) offers a new approach to enhancing goal management in MOO. The Paref methodology introduces a mathematical framework that allows users to tailor MOO algorithms to individual requirements by implementing support functions that can constrain the optimization search space in a user-defined manner. The core idea of Paref is to modify the objective function using "Pareto-reflective" functions. These functions, when concatenated with the original objective functions (black-box functions), allow the identification of Pareto-optimal points while preserving certain desirable attributes of the search space. Paref serves two main purposes in goal management for MOO:

  1. Constrain the Search Space: By introducing user-defined constraints using Pareto-reflective functions, Paref restricts the search area to focus on regions of interest. This enables the selection of Pareto-optimal solutions with specific characteristics, such as being in a particular segment or evenly distributed along the Pareto front.
  2. Guide the Optimization Process: Paref allows for the construction of MOO algorithms that adaptively reflect user preferences. For example, users can use Pareto reflections to specify constraints or properties (such as domain restrictions or targeted regions) before the optimization search. This prior definition of the search space supports more efficient goal management and decision-making, as the search process can focus directly on regions with the most potential for desired outcomes.

References:

Supervisor:  Julian Demicoli

Assigned

6. Context-Aware Adaptive Optimization: Leveraging Large Language Models in Multi-Objective Production Systems

Description: Modern dynamic systems, such as production environments, must adapt to varying conditions to achieve optimal performance across multiple objectives, including environmental impact, throughput, and reconfiguration margin in the event of failures. These adaptive systems need to continuously balance different goals, requiring complex decision-making processes often dependent on changing external factors. A practical example is a smart factory prioritizing environmental performance. During daylight, energy savings might be prioritized due to the availability of solar power. However, at night, when solar energy isn't available, the optimization strategy shifts to maintain overall environmental performance. This necessitates synthesizing high-level goals into quantifiable performance metrics without human intervention, a challenging task that demands a novel approach.

One promising solution is integrating reasoning into the system using large language models (LLMs) like GPT-4. By implementing semantic information processing, such as weather data, machine logs, and human inputs, the system can dynamically convert qualitative information into quantitative metrics. This enables more context-aware optimization in complex environments, guiding multi-objective optimization (MOO) strategies in real-time.

References:

Supervisor: Julian Demicoli

Assigned

7. Evaluation of Privacy Preserving Technologies in Healthcare: A Review of Related Work and Practical Applications

Description: In recent years, user privacy has become a critical concern, with growing recognition of the need for data sovereignty for end users. In light of this, our study aims to investigate the potential benefits of privacy-preserving technologies, including blockchain, Multi-Party Computation (MPC), and Zero-Knowledge (ZK) Proofs, specifically within critical applications such as healthcare. This exploration seeks to understand how these technologies can enhance privacy and security in managing medical data.

Tasks:

  1. Getting familiar with privacy preserving techniques.
  2. Surveying different types of data that can be shared within the medical context, and propose the best protocols/approaches for them.

References:

Supervisor: Marco Calipari

Assigned

To top