Projects

Swift advances in artificial intelligence and wireless device-to-device communications enable the edge/fog architecture as a dense wireless network consisting of intelligent entities that can learn, make decisions, and communicate. It is well-known that cooperation enables cognitive entities to subtly divide the costs, share the risks, and distribute the utility. As such, it also unlocks the true potential of edge computing in terms of resource efficiency (self-optimization), stable distributed control (self-organization), and sustainability (self-diagnosis and self-healing). Despite great potential, the implementation of cooperation in wireless networks associates with several significant hurdles. These include information shortage, heterogeneity of edge/fog nodes, communication constraints, and randomness in crucial optimization parameters. In this project, we confine our attention to three main challenges in the autonomous edge/fog computing paradigm: distributed task management, efficient resource pooling, and strategic function placement. The objective is to address these problems in a real-world system despite the abovementioned constraints by developing cooperation methods. In a nutshell, the project bridges the theory of cooperation in multi-agent systems and the practical aspects of wireless communications to address some main challenges of the edge computing paradigm. The outcomes are computationally-efficient decision-making methods that enhance the efficiency and productivity of edge computing technology concerning crucial performance metrics such as energy efficiency and service delay.

This project receives financial support from the German research foundation (DFG). The duration is 09.2023 - 08.2026.

The synergy between wireless communication, cyber-physical system design, and artificial intelligence enables the autonomous operation of modern networked systems. For such infrastructures that underpin several critical missions, the vitality of resiliency is evident and unquestionable. Nevertheless, the scarcity of resources, the inevitable implementation of technologies for opportunistic resource acquisition, and security threats, render resiliency challenging to achieve, as they introduce volatility in the essential resource flow. In this proposal, we focus on two scenarios, namely resource sharing and backup resource reservation, to boost the resilience of a mission-critical system of systems against oblivious and non-oblivious adversaries that create a volatile resource flow; As such, uncertainty and information shortage count as the focal points of our research. We maintain a generic framework of resiliency via network adaptivity so that our proposal accommodates a variety of applications. Our solutions lie at the intersection of multiagent online convex optimization with bandit feedback, online hide-and-seek games, and statistical concepts such as change point detection. The proposed methods are amenable to distributed implementation, thus reducing the feedback and signaling overhead significantly. We will provide rigorous theoretical analysis concerning efficiency, scalability, and convergence. Also, we will investigate performance bounds.

This project receives financial support from the German research foundation (DFG) in the framework of priority program (SPP) ‘Resilience in Connected Worlds”. The duration is 01.2024 - 12.2026.

Cooperation enables intelligent systems to reciprocally and subtly divide the costs, share the risks, and distribute the utility. As such, it unlocks the true potential of cognitive networks in terms of the efficiency of resource expenditure (self-optimization), stable distributed control (self-organization), and sustainability (self-diagnosing and self-healing). Cooperative cost-, risk-, and resource sharing mechanisms have broad applicability to manage and optimize technological networks. They also find application in politics and economics. Nevertheless, inducing cooperation among cognitive entities is challenging due to several issues such as lack of information, conflicting interests, and excessive computational complexity. This project aims at developing rigorous decision-making mechanisms for cooperation. Such mechanisms enable cognitive entities to strategically collaborate in complicated scenarios, e.g., under uncertainty, given communication constraints, and inside dynamic environments. The building blocks of such mechanisms are reinforcement learning or inverse reinforcement learning, together with game theory. The outcome of such mechanism are efficient equilibria.

This project receives funding from Cyber Valley as a part of the program “Cyber Valley Research Fund” and is running at the University of Tübingen. The duration is 01.2022 - 6.2025.

The use of AI/ML-based strategies that require little or no prior information increases the efficiency of technological networks in terms of several performance metrics. For example, they reduce signaling and feedback overhead, improve security and privacy, and are compatible with the self-interest of devices, vendors, and stakeholders. Nevertheless, the algorithmic decision strategies must adapt to the physical properties of the problem and the specific nature of the communication medium. For example, in wireless networks, the interference, changes in channel quality, and hardware limitations exacerbate the problem.

The core contribution of this project is the development of AI/ML-based decision strategies for application in wireless networks. Such methods shall be robust against information scarcity and amenable to distributed implementation. In particular, the goal is to apply the theoretical schemes to solve several research problems at different network layers, including radio resource management, joint communications and sensing, function placement, multi-connectivity, and beam selection.

This project is a part of the 6G-Research and Innovation Cluster. The research hub consists of several partners, including universities and research institutes. Besides, many industry partners support the project. The project receives funding from the German Ministry of Education and Research. The duration is 08.2021 - 07.2025.

Multi-armed bandit is a variety of sequential decision-making problems under uncertainty, envisaged by a gambler playing on a slot machine. While the seminal formulation comprises only one player that faces the exploration-exploitation dilemma, the challenge becomes significantly aggravated in the multi-agent setting, where the decision-makers mutually affect each other while sharing limited resources. Such scenario, which is located at the intersection of two pillars of artificial intelligence, namely, decision-making under uncertainty and multi-agent systems, requires analysis not only based on the regret performance, but also involving the concepts such as equilibrium, fairness, incentive-compatibility, revenue, and diffusion. With a forward-looking vision, the project MABISS aims at developing rigorous theoretical frameworks to address the multi-agent multi-armed bandit problem in different settings, particularly those that frequently arise in real-world applications. These include fully-distributed bandit games, bandit mechanism design, network bandits, and human bandits. Motivated with the ever-increasing demand for wireless spectrum, the application-wise focus of MABISS is the distributed intelligent spectrum sharing challenge for device-to-device communications, which is a key enabler of the emerging networking paradigms such as the Internet of Things, edge/fog computing, and small cell networks. Taking the physical characteristics of wireless networks into account, MABISS investigates the problem by practicing the theory of multi-agent multi-armed bandit and providing performance bounds. Moreover, based on the analytical and numerical results, MABISS plans to develop an intelligent spectrum sharing testbed. The application area of the results goes beyond wireless communications, ranging from science to engineering to digital health and digital humanity.

This project received funding from the German Ministry of Education and Research (BMBF) as a part of the program “promoting young female scientists in artificial intelligence”. The duration was 10.2020 - 09.2024.

Unmanned aerial vehicles (UAVs) are foreseen to significantly change the world by supplying various types of services that are complex, expensive, or dangerous to provide otherwise. In particular, UAVs assist with automation in several industrial domains such as agriculture. In the context of this project, UAVs serve as moving base-stations to enhance wireless connectivity over the agriculture site at different levels, for example, between agricultural machinery and sensors. To optimize control and resource allocation of UAV networks, artificial intelligence and machine learning are of great importance.  The core contribution of this project encompasses developing resource-efficient, accurate, and interpretable algorithms for federated- and transfer learning that can be applied by UAVs to optimize the performance of the agriculture site in different perspectives, and in particular, regarding wireless communications.

This project was a multi-party project in collaboration with several partners, including Fraunhofer Heinrich-Hertz-Institute, University of Kaiserlautern, John Deere ETIC, Welotec GmbH, and CiS GmbH. It received funding from the German Ministry of Education and Research (BMBF). The duration was 05.2020 - 10.2023.

Online learning is a variety of sequential decision-making problems under uncertainty, envisaged by an agent interacting with an unknown environment by successively selecting an action from a set of available actions. The environment can have different states. Each action, if it is selected, returns some state-dependent reward. Often, the goal of the decision-maker is to satisfy some optimality condition over the interaction horizon, which is often defined in terms of accumulated discounted reward or the accumulated regret. Graph signal processing (GSP) is a recently-established branch of signal processing. The goal of GSP is to handle the excessive amount of data that is collected in various circumstances, for example, from a sensor network, in a fast and efficient manner. In this project, the goal is to develop, adapt, and integrate the tools from graph signal processing to enable fast and efficient online learning and decision-making under uncertainty.

This project was financed by the German research foundation (DFG) in the framework of excellence cluster Machine Learning: New Perspectives for Science, at the University of Tübingen. The duration was 01.2020 - 12.2023.

The mean-field theory is a powerful tool to efficiently approximate the behavior of a complex system that includes several agents. In this approximation, the mean-field, i.e., the average effect of all agents, replaces the individual agents' interactions and serves as the basis of analysis. The mean-field theory finds applications in several domains, including network economics and optimization. The main objective of this project is to optimize the performance of ultra-dense and resource-constrained networks in different settings using the mean-field theory. The target settings include networks that involve heterogeneous agents or agents with constrained interactions.

This project was in cooperation with Dr. Marc Sedjro at AIMS South Africa, also Prof. Giuseppe Caire and Dr. Peter Jung at the Technical University of Berlin.  The project received financial support from the German Academic Exchange Service (DAAD) and German Ministry of Education and Research (BMBF). The duration was 01.2022 - 06.2024.

Despite the great potential for improving the users’ satisfaction level, network convergence is challenging, in particular with regards to control, management and planning. That includes, but is not limited to, mode- and network selection, scheduling, resource allocation, interference management, and the like. Although such challenges appear in the control and planning of almost every networking paradigm, the concept of network convergence aggravates some difficulties. In particular, each included network might have its PHY/MAC protocols, as well as a time scale. Moreover, each network has its interests due to a difference in available radio resources, subscribed users, contents (files) in high demand, and the like. In this project, we focused on self-organization for the mobile machine to machine (M2M) communications underlying a converged (broadcast-wireless-cellular) infrastructure. The problems to address include resource allocation and resource sharing, also transmission mode selection, and hand-off, taking into account the mobility and lack of information. To solve the formulated problems, we model each user as an intelligent agent and develop decision-making policies that guarantee some optimality conditions in terms of network key performance indicators and convergence. To this end, we leverage machine learning concepts. 

This project received financial support from the Einstein Center for Digital Future (ECDF) in cooperation with Deutsche Telekom AG, at the Technical University of Berlin. The duration was 05.2018 - 04.2022.

Enabled by recent technological advances, the field of radar remote sensing has entered the era of Synthetic Aperture Radar (SAR) missions with short revisit times, providing an unprecedented wealth of topography and surface change time-series. The conventional methods analyze the existing batch of radar data in an offline manner; nonetheless, with the rise of new technologies, the amount of data becomes massive in a short time, which renders offline analytics dramatically inefficient. Real-time methods, in contrast, significantly enhance the effectiveness and efficiency of interferometric SAR (InSAR) time-series analysis. Such methods are particularly useful in delay-sensitive and resource-constrained applications such as early detection of geo-hazards; Nevertheless, online analysis imposes some challenges. Specifically, it becomes imperative to develop low-complexity, easy-to-implement algorithmic solutions that adapt to the data arrivals on-the-fly. To manage the huge volume of InSAR point clouds that are obtained by multi-temporal interferometric time-series approaches, one solution is to summarize the large data sets by extracting the most important data points that represent the entire set from some specific perspective. Moreover, while analyzing the data, it is crucial to consider the temporal correlation between the data points and the features. This project focuses on developing such solutions.

This project was interdisciplinary in the framework of the HEIBRIDS Graduate School of Data Science. It was performed in cooperation with Prof. Mahdi Motagh at the German Research Center for Geosciences, Helmholtz Institute Potsdam. The duration was 2018 - 2022.

Single-cell RNA-sequencing (scRNA-seq) enables a massive acquisition of gene expressions. The measured single-cell transcription profiles can be used to identify cell types, cell sub-types and continuous gene expression gradients, e.g., during the developmental or disease processes. However, a key challenge is the high levels of sampling noise and missing data, which can obscure transcriptional measures of cell type similarity. Moreover, such impairments in the data render the identification of co-regulated groups of genes difficult. Furthermore, it is currently not possible to systematically determine the origin of cell types in complex organisms.  The focus of this project is on developing and adopting new analytical approaches to efficiently use single-cell data to solve the following problems:

• Finding informative genes that allow clustering of cells and identification of cell types
• Analysis of co-regulated gene modules
• Integration of other data types including lineage barcodes that allow tracing cell origins

To address these challenges, we propose to adapt methods from computational network science and data mining to the problem of cell type identification and to reconstruct gene regulatory networks as well as cell lineage trees. These include statistical network inference, network completion, and network extrapolation as well as tools such as dimensionality reduction, clustering, and anomaly detection.

This project was interdisciplinary in the framework of the HEIBRIDS Graduate School of Data Science. It was performed in cooperation with Prof. Uwe at the German Research Center for Geosciences, Helmholtz Institute Potsdam. The duration was 2018 - 2022.