Seminar

Driven by the requirements of innovative applications and services, the past few years have produced new technologies for networked or distributed systems. In the Internet of Things, a large number of devices and everyday objects equipped with sensors and actuators are connected to the internet and communicate through mostly wireless communication technologies (e.g., BLE, ZigBee, 6LoWPAN, LoRaWAN). In one of its sub-domains, the industrial Internet of Things (IIoT, Industry 4.0), machines, tools, transport equipment, etc. are networked.
Virtualisation (e.g., NFV) and "software-defined" systems (e.g., SDN), increase the flexiblity and efficiency of distributed systems through dynamic adaptation and scaling.
Driven by the popularity of the Bitcoin system, distributed ledger technologies and related concepts such as smart contracts have been developed, which are not only the foundations of electronic currencies, but can also support any application in which a consensus between different parties must be reached and documented. 
Another focus has been the reduction of latency in networked and distributed systems, e.g. by using nearby edge and fog computing resources in addition to the remote cloud, or by using optimised communication protocols to., for instance, rapidly connect client and server. 
In addition to stationary networks, mobile (5G) communication technologies and systems have developed rapidly. For example, the Covid-19 tracing application uses mobile devices to track contacts. This method, known as crowdsensing, can be used more generally to collect large amounts of geographically distributed sensor data. 

This seminar will discuss a wide range of current technologies, protocols and standards that enable the above networked and distributed applications and services.

Organization: The seminar is organized in the style of a scientific conference. Following the submission of a written paper on the assigned topic, students write reviews for other seminar papers and participate in a final presentation session where they present their work and discuss the work of others. Attendance at the kick-off and final presentation session is mandatory.

Formal prerequisites: Successful completion of at least 1 course at the department of Distributed Systems.

Time-Sensitive Networks

Topic 1: Deterministic Real-time communication with Time-sensitive Networking

Supervisor: Lucas Haug

Real-time communication with deterministic bounds on network delay and delay variance (jitter) is crucial for the efficient and safe operation of networked real-time systems. As distributed Cyber-Physical Systems (CPSs) with networked sensors, actuators, and controllers become increasingly popular in various domains including the Industrial Internet of Things (IIoT) and autonomous vehicles, the demand for deterministic real-time communication that provides high reliability and bounded network delay and jitter has grown.

Major standardization organizations like the Institute of Electrical and Electronics Engineers (IEEE) and the Internet Engineering Taskforce (IETF) have acknowledged the necessity for deterministic networks, leading to a set of standards that are subsumed under the term Time-Sensitive Networking (TSN). These standards extend traditional wired IEEE 803.3 networks (Ethernet) by real-time communication mechanisms. For example, IEEE 802.1Qbv defines the Time-Aware Shaper, which enables deterministic scheduling of traffic through time-slotting. Similarly, IEEE 802.1Qbu and IEEE 802.3br introduce Frame Preemption to allow high-priority traffic to interrupt lower-priority transmissions. IEEE 802.1CB introduces Frame Replication and Elimination for Reliability (FRER).

In this seminar topic, an overview of the aforementioned TSN standards shall be given. Additionally, it shall be described how they enable deterministic real-time communication.

Networking Technologies

Topic 2: Media Access Control in IEEE 802.11 Networks

Supervisor: Jona Herrmann

Wireless Local Area Networks (WLANs) based on the IEEE 802.11 standard are very important for mobile devices, as this is the only way they can access the network and, finally, the Internet. Typically, many devices are connected to the same WLAN and use it at the same time. Therefore, a Media Access Control (MAC) protocol is required to coordinate access to the shared medium.

In this seminar topic, we will explore why the Media Access Control of IEEE 802.3 networks cannot be used in IEEE 802.11 networks. Additionally, the seminar paper should explain the MAC protocol that is used instead.

Topic 3: Evolution of the IEEE 802.11 Standard

Supervisor: Jona Herrmann

Wireless Local Area Networks (WLANs) based on the IEEE 802.11 standard, have become an integral part of modern communication and are now available almost everywhere, including workplaces, homes, public places, and public transportation. This standard is constantly evolving in order to provide better data rates, latency, and energy efficiency. These developments have resulted in improved standard versions, such as the most recent IEEE 802.11be.

In this seminar topic, we will compare the different IEEE 802.11 versions and explain their key concepts.

Topic 4: QUIC -The fast TCP alternative

Supervisor: Michael Schramm

For large web companies and Content Distribution Networks (CDN), the performance and security of web protocols is a major concern. For this reason, Google proposed the QUIC protocol. It is supported by all Google mobile apps and as of October 2020, more than 75% of Meta Internet traffic used QUIC, making it Meta's standard Internet protocol [1]. But how does this novel protocol ensure security? How is the higher performance achieved and how much higher is it? What is the Packet Pacing, which is used in QUIC and why can it be superior to techniques used in TCP (e.g. Fast Retransmit)?

In this seminar topic, we will explain the QUIC protocol and compare it to TCP protocol variants. Additionally, we will review scientific papers to extract performance measurements and experiences with the QUIC protocol ^[source].

Topic 5: Tor - The Onion Router

Supervisor: Lukas Epple

Tor is a popular anonymity network that allows users to browse the Internet without revealing their true identity, location, or activity. By routing Internet traffic through a series of encrypted nodes or relays, the source of the traffic is obscured and user privacy is maintained.

In this seminar topic, we will gain a comprehensive overview of Tor and its technical underpinnings. This will be done by exploring how Tor works internally and how it achieves privacy. Additionally, the seminar paper should review the potential risks and limitations of Tor, as well as its impact on Internet privacy and security.

Topic 6: Extended Berkeley Packet Filters (eBPF) and eXpress Data Path (XDP)

Supervisor: Simon Egger

Many applications require high-frequency traffic at 10 Gbps or higher, but the socket interface and networking stack of the operating system are often a bottleneck. Kernel-bypass and Berkley Packet Filter (BPF) therefore specifically target high throughput. Kernel-bypass allows applications to skip the kernel network stack and receive packets directly from the network card or driver. BPF performs the processing (e.g., packet filtering) directly on a virtual machine in the kernel space, which improves speed and efficiency compared to user-space alternatives and avoids context switches. The Linux networking feature eXpress Data Paths (XDP), built on the extended Berkley Packet Filter (eBPF), combines both methods.

In this seminar topic, we will first investigate the use cases and motivators that raise the demand for high throughput. We will then investigate why standard BPF are relatively slow, followed by an overview of XDP and eBPF. While eBPF is a very powerful technology that is also used in other application areas (e.g., performance analysis of kernel functions), the focus of the seminar work should be on fast network communication with XDP. Additionally, a comparison with related technologies such as netmap (FreeBSD) and PF_RING, addressing both similarities and differences, should be performed.

• [1] T. Høiland-Jørgensen et al.: “The eXpress data path: fast programmable packet processing in the operating system kernel”. Proceedings of the 14th International Conference on emerging Networking EXperiments and Technologies (CoNEXT 2018), 2018, DOI: 10.1145/3281411.3281443.
• [2] M. A. M. Vieira et al.: “Fast Packet Processing with eBPF and XDP: Concepts, Code, Challenges, and Applications”. ACM Computing Surveys, 53(1), 2020, DOI: 10.1145/3371038.
• [3] L. Rizzo: netmap: “A novel framework for fast packet I/O”. Proceedings of the 2012 USENIX conference on Annual Technical Conference (USENIX ATC 2012), 2012.
• [4] L. Deri: “Modern packet capture and analysis: Multi-core, multi-gigabit, and beyond”. Tutoria at the 11th IFIP/IEEE International Symposium on Integrated Network Management (IM), 2009, http://luca.ntop.org/IM2009_Tutorial.pdf

Concurrent Programs

Topic 7: Memory Ordering in the C++ Standard

Supervisor: Simon König

Controlling access to memory plays a crucial role when building concurrent programs. Programmers often assume the operations of their programs to be executed sequentially and in-order. However, optimizations applied by the compiler, the CPU pipeline and the memory management unit can cause the operations of a program to be executed out-of-order. In multi-threaded applications, this has the potential to change the semantics of the program, resulting in unexpected observed behavior. Therefore, synchronization is necessary. The usage of memory barriers allows OS-level primitives (e.g., semaphores) or lock-free data structures to provide meaningful consistency guarantees.

In this seminar topic, we will delve into the principles and mechanisms that are used to control and synchronize access to shared memory. Specifically, we will investigate the usage and semantics of memory barriers in modern memory models such as the C++ memory model. Hereby, a special focus will be on the theoretical model behind acquire-release or sequentially consistent memory ordering as defined by the C++ standard.

Topic 8: User-Space Mutex: Spinlocks and Atomic Synchronization

Supervisor: Simon König

This seminar topic focuses on the challenges of ensuring proper synchronization without relying on kernel-based synchronization. We will explore the concept of user-space mutex implementations, focusing particularly on spinlocks and their role in low-level synchronization within concurrent systems. We will analyze the mechanics of busy-wait locks, where threads continuously check the availability of a critical section, and the implications this has on processor utilization and cache coherence.

The discussion will center on spinlock implementations as synchronization primitives that avoid blocking and instead rely on active polling. Specifically, we will compare the test-and-set lock, the test-and-test-and-set lock, and MCS locks. The focus shall lie on the trade-offs in terms of efficiency under contention and multithreaded scalability.

IoT

Topic 9: Reference Architectures for Vehicle Testing

Supervisor: Melanie Heck

Software plays an increasingly important role in vehicles, but many components are inherently uncertain and must be able to survive in highly dynamic environments. With smart technologies such as autonomous driving and smart roadside infrastructure, such dynamic environments are becoming more present and increasingly challenging. Therefore, continuous testing of vehicle components, especially after an update, is essential to ensure that the vehicle still functions correctly in any possible scenario. This requires architectures that have been designed specifically for testability and support test activities as efficiently as possible. However, it would be highly inefficient to design a new architecture from scratch for each vehicle.  This calls for new reference architectures that reuse architectural knowledge in the form of architectural styles and patterns for the same problem classes.

In this seminar topic, we will review and compare different vehicle testing reference architectures that have been proposed in the literature. 

Melanie Heck

The Internet of Everything (IoE), where virtually everything can now communicate through the Internet, and the increasingly demanding performance requirements of new technologies (e.g., cryptocurrencies) have driven the emergence of new computing paradigms for distributed systems. Scalability is now offered not only by centralized cloud providers, but also by edge computing systems, where geographically distributed servers provide computational resources at the edge of the network and, therefore, close to the end devices. This can significantly reduce latency for time-critical applications like vehicular networks. The advances in edge computing have led to the emergence of edge AI, where powerful AI algorithms are deployed at the edge, without relying on a remote cloud.

But distributed systems come with many challenges which requires a profound understanding of core principles in distributed computing. As pointed out by former Google Senior Vice President Urs Hölzl: “At scale, everything breaks ... Keeping things simple and yet scalable is actually the biggest challenge. It's really, really hard.“ This is especially true for dynamic and uncertain environments that we are facing, for instance, in smart buildings or smart energy systems. Self-adaptation is one of the key mechanisms for coping with increasingly large and dynamic systems, often by using machine learning techniques (GNN, reinforcement learning). Challenges that come with distributed storage systems include consistency and scalability.

Another hot topic, especially in the context of 5G and the development of future 6G networks, is Time Sensitive Networking (TSN), which defines a set of standards to enable reliable, deterministic real-time communication in Ethernet networks. These standards target, among others, time synchronization and traffic shaping/scheduling approaches for both event-based and time-triggered traffic.

In this seminar, we take a deep dive into specific concepts of distributed and context-aware systems that tackle the above challenges.  The topics will be published on the department’s website and are assigned according to a standardized procedure as explained during the kick-off.

Organization: The seminar is organized in the style of a scientific conference. Following the submission of a written paper on the assigned topic, students write reviews for other seminar papers and participate in a final presentation session where they present their work and discuss the work of others. Attendance at the kick-off and final presentation session is mandatory.

Formal prerequisite: Successful completion of at least 1 Master-level course at the department of Distributed Systems.

Efficient and Reliable Network Communication

Topic 1: Evolution of the IEEE 802.11 Standard

Supervisor: Jona Herrmann

Wireless Local Area Networks (WLANs) based on the IEEE 802.11 standard, have become an integral part of modern communication and are now available almost everywhere, including workplaces, homes, public places, and public transportation. This standard is constantly evolving in order to provide better data rates, latency, and energy efficiency. These developments have resulted in improved standard versions, such as the most recent IEEE 802.11be.

In this seminar topic, we will compare the different IEEE 802.11 versions and explain their key concepts.

Topic 2: Connection Establishment and Roaming in IEEE 802.11 Networks

Supervisor: Jona Herrmann

In order to use a Wireless Local Area Network (WLAN), mobile devices must first connect to it. The authentication and association procedure is defined in the IEEE 802.11 standard. To provide wide coverage, WLANs typically have several Access Points (APs). Therefore, to ensure seamless connectivity during mobility, roaming is required.

In this seminar topic, we will examine the principles and mechanisms behind connection establishment and roaming in IEEE 802.11 networks.

Topic 3: gPTP-based Time Synchronization in Converged 6G/TSN networks

Supervisor: Lucas Haug

Accurate clock synchronization is essential in networks that require time-critical data transfer, where even minimal clock deviations can lead to significant operational errors. Given that the internal clocks of network devices inherently drift apart over time, clock synchronization is indispensable for reliable and error-free operations.

In wired Ethernet networks, protocols like the Generalized Precision Time Protocol (gPTP), defined in the IEEE 802.1AS standard, are widely implemented. gPTP employs a master-slave architecture and hardware timestamping to achieve time synchronization with sub-microsecond accuracy. However, in emerging network domains – including integrated 6G/TSN networks – further problems, such as variable link delays, arise. Thus, time synchronization in these fields is still under active research.

In this seminar topic, we will investigate the mechanisms through which gPTP achieves precise time synchronization in traditional Ethernet-networks and how they can be employed in integrated 6G/TSN networks.

Topic 4: Radio Resource Grid Allocation for 5G Network Slicing

Supervisor: Lucas Haug

Efficient allocation of radio resources is critical in 5G networks, where diverse services such as enhanced mobile broadband (eMBB), ultra-reliable low-latency communications (URLLC), and massive machine-type communication (mMTC) must coexist. Network slicing enables multiple logical networks to share the same physical infrastructure while maintaining performance isolation. However, realizing slicing at the Radio Access Network (RAN) level introduces challenges, particularly in the allocation and scheduling of radio resources within the resource grid. Ensuring that different slices operate efficiently while minimizing interference and maximizing spectral efficiency is a central research problem.

The resource grid in 5G is a structured framework that defines how the radio spectrum is divided into time-frequency units. In RAN slicing, each slice may require a different configuration in terms of bandwidth, latency, and reliability, necessitating dynamic partitioning of the resource grid. This involves defining slice-specific configurations at different protocol layers, such as adaptive scheduling at the Medium Access Control (MAC) layer and customized numerologies at the physical layer (PHY). Furthermore, ensuring isolation between slices while maintaining overall network efficiency requires sophisticated resource management strategies, including dynamic spectrum allocation and interference mitigation techniques.

In this seminar topic, we will examine how the resource grid in 5G RAN slicing is structured and allocated across different slices with a possible focus on resource scheduling, isolation mechanisms, or strategies for efficient spectrum utilization.

Topic 5: An Introduction to Semantic Communication

Supervisor: Simon Egger

Semantic communication is an emerging research trend that aims to overcome the Shannon limit of traditional encoding and modulation schemes.

In this seminar topic, we will provide an overview of this research trend, starting from Weaver's famous categorization of communication problems [1] into: (A) How accurately can the symbols be transmitted? (B) How precisely do the transmitted symbols convey the desired meaning? (C) How effectively does the received meaning affect conduct in the desired way? As this categorization was mainly focused on human-to-human communication, we will also investigate more recent semantic channel models, attack/failure models, and their effectiveness/applicability in modern applications (e.g., [1], [2] for initial references).

• [1] W. Weaver and C. Shannon, “Recent contributions to the mathematical theory of communication,” Mathematical Theory Commun., University of Illinois Press, 1949.
• [2] Luo, Xuewen, Hsiao-Hwa Chen, and Qing Guo. "Semantic communications: Overview, open issues, and future research directions." IEEE Wireless communications 29.1 (2022): 210-219.
• [3] B. Güler, A. Yener and A. Swami, "The Semantic Communication Game," in IEEE Transactions on Cognitive Communications and Networking, vol. 4, no. 4, pp. 787-802, Dec. 2018.

Topic 6: Highly-Available Control Planes in Software-Defined Networking

Supervisor: Simon Egger

Software-Defined Networking enables dynamic network management and network programmability by decoupling the network control and data plane. This is achieved through a network controller that can provide both low-level control (e.g., adding/removing a switch) and high-level control (e.g., load-balancing) over the network. While logically centralized, modern SDN controllers often follow an microservice-based design (e.g., Google's Orion [1]) that splits the controller independent modules for routing decisions.

In this seminar topic, we will look at the key challenges of consistency and availability of network control functions in this setting. We will investigate the design choices of Orion [2], consistency models to enable network updates [2], and recent approaches on formally verified high-availability control planes [3].

• [1] Ferguson, Andrew D., et al. "Orion: Google's Software-Defined Networking Control Plane." 18th USENIX Symposium on Networked Systems Design and Implementation (NSDI 21). 2021.
• [2] Reitblatt, Mark, et al. "Abstractions for network update." ACM SIGCOMM Computer Communication Review 42.4 (2012): 323-334.
• [3] Namyar, Pooria, et al. "ZENITH: Towards A Formally Verified Highly-Available Control Plane." Proceedings of the ACM SIGCOMM 2025 Conference. 2025.

Topic 7: Utreexo

Supervisor: Lukas Epple

Utreexo offers a pathway to minimize Bitcoin’s Unspent Transaction Output (UTXO) set storage through the use of cryptographic accumulators. This enables the operation of lightweight, fully validating nodes using only a few kilobytes per node while maintaining network security and integrity.

In this seminar topic, we will gain an overview of Utreexo and its transformative potential by examining its impact on node management, decentralization, and the future prospects of blockchain technology. To this end, the seminar paper should explore the internal workings of Utreexo and its potential to improve scalability and participation in the Bitcoin network. 

Concurrent Programs

Topic 8: Memory Ordering in the C++ Standard

Supervisor: Simon König

Controlling access to memory plays a crucial role when building concurrent programs. Programmers often assume the operations of their programs to be executed sequentially and in-order. However, optimizations applied by the compiler, the CPU pipeline and the memory management unit can cause the operations of a program to be executed out-of-order. In multi-threaded applications, this has the potential to change the semantics of the program, resulting in unexpected observed behavior. Therefore, synchronization is necessary. The usage of memory barriers allows OS-level primitives (e.g., semaphores) or lock-free data structures to provide meaningful consistency guarantees.

In this seminar topic, we will delve into the principles and mechanisms that are used to control and synchronize access to shared memory. Specifically, we will investigate the usage and semantics of memory barriers in modern memory models such as the C++ memory model. Hereby, a special focus will be on the theoretical model behind acquire-release or sequentially consistent memory ordering as defined by the C++ standard.

Topic 9: User-Space Mutex: Spinlocks and Atomic Synchronization

Supervisor: Simon König

This seminar topic focuses on the challenges of ensuring proper synchronization without relying on kernel-based synchronization. We will explore the concept of user-space mutex implementations, focusing particularly on spinlocks and their role in low-level synchronization within concurrent systems. We will analyze the mechanics of busy-wait locks, where threads continuously check the availability of a critical section, and the implications this has on processor utilization and cache coherence.

The discussion will center on spinlock implementations as synchronization primitives that avoid blocking and instead rely on active polling. Specifically, we will compare the test-and-set lock, the test-and-test-and-set lock, and MCS locks. The focus shall lie on the trade-offs in terms of efficiency under contention and multithreaded scalability.

Topic 10: Coroutines - A new form of Subroutines

Supervisor: Lukas Epple

Subroutines encapsulate some computation and help to break down a large program into smaller parts. Coroutines imitate this property, but the lifetime of a coroutine is not tied to the control flow of the program. When a subroutine returns to the calling program, its control information and local variables are destroyed. In contrast, when a coroutine returns control to its caller, its execution is not finished and thus its state is preserved. Each time control reenters the coroutine, it resumes execution where its local control left off and data state is retained. Hence, coroutines are a form of retentive control.

Coroutines are especially well suited for the implementation of asynchronous or event-driven applications. With the use of coroutines, data or IO stalls can be hidden, improving parallelism at runtime and reducing programming complexity at build-time. For this reason, coroutines are experiencing a resurgence. Many popular languages have recently introduced native support for coroutines. Moreover, coroutine-based implementations of highly parallel applications are becoming more and more popular.

In this seminar topic, we will examine the principles and mechanisms of coroutines. Through a comprehensive review of relevant literature and case studies, the seminar paper could investigate aspects including synchronization primitives for coroutines, state and context switch mechanics for coroutines, and mechanisms for runtime optimization.

Self-Adaptive Systems

Topic 11: Algorithms for Predicting Smart Home Behavior: Enabling Flexible and Energy-Efficient Living

Supervisor: Justin Schiel

The increasing adoption of smart home technologies introduces new opportunities for energy-efficient living. A key challenge is predicting and managing smart home behavior in ways that leverage the flexibility of user-acceptable states. For example, an occupant may be comfortable with indoor temperatures ranging between 20°C and 23°C. This tolerance can be exploited to optimize energy consumption without sacrificing comfort. Algorithms that predict smart home behavior are central to enabling such flexible energy management.

In this seminar topic, we will provide an overview of various algorithms for predicting smart home behavior, including statistical models, optimization-based approaches, and machine learning techniques. Unlike traditional rule-based methods, these algorithms provide more sophisticated means of capturing system dynamics. The seminar paper will focus on understanding how these algorithms operate, including the underlying principles, data requirements, and decision-making processes.

Topic 12: Reference Architectures for Vehicle Testing

Supervisor: Melanie Heck

Software plays an increasingly important role in vehicles, but many components are inherently uncertain and must be able to survive in highly dynamic environments. With smart technologies such as autonomous driving and smart roadside infrastructure, such dynamic environments are becoming more present and increasingly challenging. Therefore, continuous testing of vehicle components, especially after an update, is essential to ensure that the vehicle still functions correctly in any possible scenario. This requires architectures that have been designed specifically for testability and support test activities as efficiently as possible. However, it would be highly inefficient to design a new architecture from scratch for each vehicle.  This calls for new reference architectures that reuse architectural knowledge in the form of architectural styles and patterns for the same problem classes.

In this seminar topic, we will review and compare different vehicle testing reference architectures that have been proposed in the literature.

Topic 13: Gaze Fixation and Saccade Detection Algorithms

Supervisor: Melanie Heck

A person’s eye movements not only reveal what objects they are looking at, but can also be used to derive emotions, stress levels, mental load, and other cognitive states. This information can then in turn be used to dynamically adapt applications in a way that better supports the user and their current state of mind. However, the raw gaze data is often too noisy and volatile to make any meaningful inferences. Sophisticated algorithms have therefore been developed that calculate two basic gaze features: fixations and saccades. While fixations group gaze points during which a person focuses at a specific object, saccades mark the transitions between two such focus points. Depending on the sampling rate and quality of the collected gaze data, different algorithms can be used to detect fixations and saccades.

In this seminar topic, we will review the most commonly used fixation and saccade detection algorithms and, based on a discussion of their respective advantages and shortcomings, analyze their suitability for different eye tracking datasets.

Self-Adaptive Systems

Topic 14: Federated Learning for the Edge

Supervisor: Michael Schramm

Federated Learning is a machine learning setting where many clients train together without making the data itself accessible to collaborators. This enables institutions with smaller datasets to gain insights that they could not get from their own data alone. In addition, federated learning can help in privacy-sensitive areas such as medical research.

At the same time, edge computing has attracted the interest of researchers and engineers to accelerate content delivery and reduce network load by communicating with nearby edge nodes instead of the cloud. However, it introduces new issues such as task scheduling, data replication, and data placement.

In this seminar topic, we will review scientific publications on federated learning for the edge. What are the challenges in edge networks? What federated learning approaches exist to tackle these challenges?

Topic 15: AI for the Edge

Supervisor: Michael Schramm

With the evolution of mobile communication technologies, edge computing theory and techniques have increasingly attracted the interest of researchers and engineers around the world. Edge computing can help to accelerate content delivery and reduce network load by communicating with nearby edge nodes instead of the cloud.

Artificial Intelligence (AI) is increasingly used to improve the performance, scalability, and efficiency of edge computing environments. AI techniques are applied for dynamic resource allocation, predictive maintenance, and real-time workload distribution. By leveraging machine learning and deep learning models, edge networks can self-optimize, adapt to changing conditions, and reduce operation costs.

In this seminar topic, we will explore how AI can optimize edge computing environments. This will be done by investigating how AI can be used for resource management, network optimization, security, and energy efficiency at the edge. Additionally, the seminar paper should discuss the benefits and limitations of AI-driven edge optimization and examine the trade-offs between computational overhead and performance gains.

Melanie Heck