Announcement

April 2023 - APPLY FOR MY NEW PHD POSITIONS or MSC PROJECTS

Current Projects

Novel approaches to Internet congestion control protocol design

Internet congestion control is a critical factor affecting both network efficiency and user experience. In recent years, there has been a growing interest in exploring novel approaches to Internet congestion control protocol design. Some of the new design objectives include developing application-oriented and quality of experience (QoE)-aware protocols. Researchers are also experimenting with new algorithmic approaches, such as control-theoretic and learning-based methods, to improve the efficiency of congestion control. In addition, rethinking the design space has led to the development of novel schemes for multipath congestion control, which can effectively mitigate congestion across multiple network paths. These novel approaches have the potential to significantly improve the performance of Internet congestion control protocols and enhance the user experience. The primary objective of this PhD research project is to explore innovative and practical concepts related to Internet congestion control, develop and analyze new algorithms optimized for new aspects mentioned above.

Congestion management, SLA enforcement and better quality of experiences on the Internet and private network domains

Despite extensive research and standardization in the area of Quality of Service (QoS), most of the developed solutions have not been deployed in practice. Proponents of overprovisioning argue that it is much easier and more efficient to add capacity when needed than to build and maintain complex QoS mechanisms that only provide minor improvement during congestion. Network congestion together with the ways of avoiding its impacts on the end users is a recurrent aspect along time, especially nowadays where the network is becoming a critical societal asset. Recently, the unfortunate pandemic situation due to COVID-19 made evident this critical fact, revealing the need of defining proper mechanisms for improving its robustness and ensuring sufficient quality of experience for end users. To improve the QoS and quality of experience (QoE) of end users different congestion control algorithms (CCAs) have been developed. For long, they were required to be fair with each other ont he public Internet. However, ensuring fairness among a large variety of CCAs is generally impossible and blocks innovation in this area. Bob Briscoe dismantled the religion of perflow-fairness few years ago by pointing to that flows were not the economic actors on the Internet, and thus there was no reason to treat them equally (or in some weighted fashion). Briscoe’s argument is that congestion control determines, to some extent, the allocation of bandwidth on the public Internet; thus, these allocations should be motivated by some underlying economic model, and per-flow fairness had no such intellectual foundation. Scott Shenker and his team also investigated this topic and proposed the concept of Recursive Congestion Shares (RCS) to capture the economic nature of resource sharing ont he Internet. They identified key issues to be addressed: Where is the isolation enforced? How are congestion-shares computed? What happens within originating and terminating domains? The initial implemetation of RCS does not deal with several important aspects like QoS, SLA, traffic engineering and incremental deployability, posing many open questions to be addressed during the course of this PhD research project.

Moving target defenses and other in-network security solutions in programmable networks

Moving Target Defense (MTD) has emerged as a solution that provides proactive defense against adaptive adversaries. The goal of MTD is to constantly move between multiple configurations in a cyber-system (such as changing the open network ports, network configuration, software, etc..) thereby increasing the uncertainty for the attacker; in effect, diminishing the advantage of reconnaissance that an attacker inherently has against traditional defense mechanisms. The advantages of MTDs go away if the shifting mechanism is deterministic because the attacker, with time on their side, will eventually be able to predict this movement and design attacks accordingly. Thus, for MTDs to be effective, they need to have implicit randomness built into them. This survey categorizes MTDs based on what they shift, when they shift and how they shift. The dynamic aspect of MTD adds an extra layer of complexity in implementing these defenses. To address this, programmable networks including SDN and data plane programmability like P4 can introduce new opportunities to realize MTD. Key challenges to be addressed during the PhD research: 1) configuration set (what to move?), 2) timing (when to move?), 3) way of movement (how to move?), 4) evaluation models and 5) new MTD approaches based on stochastic modelling or the applicationl of game theory.

Explainability methods for AI/ML models in different application areas

AI and ML models have gained a siginificant attention in the past years. In many areas, deep learning (DL) approaches proved to be accurate prediction and modelling tools. However, the black-box nature of these models challenges their use in mission critical applications, raising ethical and judicial concerns inducing lack of trust. Explainable Artificial Intelligence (XAI) is a field of AI that promotes a set of tools, techniques, and algorithms that can generate high-quality interpretable, intuitive, human-understandable explanations of AI decisions. Explainability methods for computer vision models have widely been analyzed in the past. However,the explainability of other DL models like Graph Neural Networks has gained less attention and the existing methods perform poorly on real data sets. In this PhD research, we will mostly focus on graph neural network models and their explainability in selected application areas. Application areas can determine the evaluation methodology of explainability feature and can also provide additional information (e.g., given by an area expert) that can be incorporated into the models. The goal of this research project is to create explainable models for the various problems of the selected application areas. One candidate area in this thesis project is the optimization of mobile cellular networks. In mobile cellular networks advanced network observability and analytics is essential for high quality network management. In network performance management, getting meaningful insights is essential for the network operation to be able to perform (either manually or automatically) the proper actions when necessary. Key Performance Indicators (KPIs) are used as the main descriptors of network performance. In case of low KPI value, it is important to get proper diagnosis to the problem. Root-cause analysis is the process of identifying the main source of the performance degradation. After finding the root cause the operator can potentially take further actions to fix the problem or improve the network to avoid similar problems in the future.