Security assessments of networked systems require knowledge about the utilized communication protocol. For proprietary protocols without known specification and with only limited access to the end-points, the only source of information is the communication itself. To correctly conclude from the captured byte stream to message-formats, -types, and finally a protocol model, structure, message- and field-boundaries, data-type, and semantics need to be inferred.After an initial inference procedure, it is desirable to refine the existing protocol model. Additional information gained by recorded network traffic needs to be incorporated by recognizing the appropriate parts of the model. The modeled knowledge is to be extended depending on the applicable information inferable from the new trace.

To build powerful machine learning models, lots of data is required. However, obtaining the data comes with privacy risks for the people or entities that provide their data. Recently, Google published TensorFlow Federated - an open source framework to allow machine learning on decentralized data. The approach of federated learning makes machine learning in the age of mobile devices and wearables both more efficient, as well as more privacy-friendly. The goal of this thesis or project is to become familiar with the TensorFlow Federated framework, to understand and be able to explain the techniques which are implemented in it, to be able to build machine learning models in a federated way, and possibly to implement own enhancements of the framework.

Machine learning offers great opportunities, but also comes with risks. Especially the privacy risks are becoming more prevalent in the discussions about machine learning. Recently, Google published a machine learning library called TensorFlow Privacy. Its goal is to make it easier for developers and researchers to build privacy-preserving machine learning models. Specifically, it utilizes Differential Privacy, which mathematically guarantees that the training data to create the models is protected from being extracted. The goal of this thesis or project is to become familiar with the TensorFlow Privacy library, to understand and be able to explain the techniques which are implemented in it, to be able to build privacy-preserved machine learning models, and possibly to implement own protection techniques that could enhance the TensorFlow Privacy library.

Security assessments of networked systems require knowledge about the utilized communication protocol. For proprietary protocols without known specification and with only limited access to the end-points, the only source of information is the communication itself. To correctly conclude from the captured byte stream to message-formats, -types, and finally a protocol model, structure, message- and field-boundaries, data-type, and semantics need to be inferred.After an initial inference procedure, it is desirable to refine the existing protocol model. Actively probing an entity for the validity of message syntaxes allows to targetedly enhance the knowledge about the protocol. To do this efficiently a smart method of automatically generating test-cases depending on the current protocol model needs to be developed.

Modern vehicles will use communication to increase the safety of its passengers, reduce fuel consumption, travel time, and more. The communication between the vehicles will be mainly beacon messages containing the speed, position, acceleration and other properties. These messages need to be validated, if they contain correct (plausible) information. For example, when a vehicle is suddenly stopping, but sending an increase in speed, the following vehicles may crash into the misbehaving vehicle. In literature, there is already existing work on detecting misbehavior in the data with different techniques such as subjective logic or machine learning. In this project, we will analyze the VeReMi data-set with the help of different machine learning algorithms. The number of algorithms compared is depending on the scope (credits). The student can choose the framework, e.g. PyTorch.

Encryption is one of the most reliable techniques for protecting information. However, once data is encrypted, using it becomes very difficult. Goal of this thesis or project, is to explore how Machine Learning algorithms can be designed to be able to deal with encrypted data. Firstly, a survey of existing mechanisms should be conducted. In a second part, algorithms will be comparatively implemented, or own encryption mechanisms introduced.

Security assessments of networked systems require knowledge about the utilized communication protocol. For proprietary protocols without known specification and with only limited access to the end-points, the only source of information is the communication itself. To correctly conclude from the captured byte stream to message-formats, -types, and finally a protocol model, structure, message- and field-boundaries, data-type, and semantics need to be inferred.After an initial inference procedure, it is desirable to refine the existing protocol model. Additional information gained by recorded network traffic needs to be incorporated by recognizing the appropriate parts of the model. The modeled knowledge is to be extended depending on the applicable information inferable from the new trace.

To build powerful machine learning models, lots of data is required. However, obtaining the data comes with privacy risks for the people or entities that provide their data. Recently, Google published TensorFlow Federated - an open source framework to allow machine learning on decentralized data. The approach of federated learning makes machine learning in the age of mobile devices and wearables both more efficient, as well as more privacy-friendly. The goal of this thesis or project is to become familiar with the TensorFlow Federated framework, to understand and be able to explain the techniques which are implemented in it, to be able to build machine learning models in a federated way, and possibly to implement own enhancements of the framework.

Machine learning offers great opportunities, but also comes with risks. Especially the privacy risks are becoming more prevalent in the discussions about machine learning. Recently, Google published a machine learning library called TensorFlow Privacy. Its goal is to make it easier for developers and researchers to build privacy-preserving machine learning models. Specifically, it utilizes Differential Privacy, which mathematically guarantees that the training data to create the models is protected from being extracted. The goal of this thesis or project is to become familiar with the TensorFlow Privacy library, to understand and be able to explain the techniques which are implemented in it, to be able to build privacy-preserved machine learning models, and possibly to implement own protection techniques that could enhance the TensorFlow Privacy library.

Security assessments of networked systems require knowledge about the utilized communication protocol. For proprietary protocols without known specification and with only limited access to the end-points, the only source of information is the communication itself. To correctly conclude from the captured byte stream to message-formats, -types, and finally a protocol model, structure, message- and field-boundaries, data-type, and semantics need to be inferred.After an initial inference procedure, it is desirable to refine the existing protocol model. Actively probing an entity for the validity of message syntaxes allows to targetedly enhance the knowledge about the protocol. To do this efficiently a smart method of automatically generating test-cases depending on the current protocol model needs to be developed.

Modern vehicles will use communication to increase the safety of its passengers, reduce fuel consumption, travel time, and more. The communication between the vehicles will be mainly beacon messages containing the speed, position, acceleration and other properties. These messages need to be validated, if they contain correct (plausible) information. For example, when a vehicle is suddenly stopping, but sending an increase in speed, the following vehicles may crash into the misbehaving vehicle. In literature, there is already existing work on detecting misbehavior in the data with different techniques such as subjective logic or machine learning. In this project, we will analyze the VeReMi data-set with the help of different machine learning algorithms. The number of algorithms compared is depending on the scope (credits). The student can choose the framework, e.g. PyTorch.

Neurofeedback provides the necessary means to visualize selected and controlled parameters of the brain activity. In healthcare domain, neurofeedback studies enable mitigation of many psychological disorders and illnesses, mainly by therapies that help patients to better self-regulate their brain activity. Electroencephalography (EEG) is the method of monitoring the electrical activity of the brain, thus providing the necessary feedback. In this thesis work, the student is required to survey the current state of frameworks, techniques, or methods that enable coupling of Mobile EEGs with Smartphones. Bluetooth 2.1 with Enhanced Data Rate (EDR) capability is one of the most effective mean of coupling EEGs with Smartphones. The student would therefore be required to work on the Bluetooth stack to acquire real-time data generated from the Mobile EEGs, parse the electrical signal, and visualize the signal semantically. For successful completion of the thesis, the student would be required to identify and address any one of the open challenges faced by the proposed topic. An example of this can be addressing the bandwidth challenges, battery consumption, or signal accuracy

Encryption is one of the most reliable techniques for protecting information. However, once data is encrypted, using it becomes very difficult. Goal of this thesis or project, is to explore how Machine Learning algorithms can be designed to be able to deal with encrypted data. Firstly, a survey of existing mechanisms should be conducted. In a second part, algorithms will be comparatively implemented, or own encryption mechanisms introduced.

Security assessments of networked systems require knowledge about the utilized communication protocol. For proprietary protocols without known specification and with only limited access to the end-points, the only source of information is the communication itself. To correctly conclude from the captured byte stream to message-formats, -types, and finally a protocol model, structure, message- and field-boundaries, data-type, and semantics need to be inferred.After an initial inference procedure, it is desirable to refine the existing protocol model. Additional information gained by recorded network traffic needs to be incorporated by recognizing the appropriate parts of the model. The modeled knowledge is to be extended depending on the applicable information inferable from the new trace.

Die Topologie von Peer-to-Peer-Netzen spielt für viele darauf aufbauende Protokolle eine zentrale Bedeutung. So bestimmt der Netzwerkdurchmesser beispielsweise, wie schnell alle Teilnehmer Broadcastnachrichten erhalten können. Zudem agieren Teilnehmer in einem Peer-to-Peer-Netzwerk üblicherweise so, dass sie ihre ei-gene Situation verbessern. Basierend darauf lassen sich die Hand-lungen der Teilnehmer spieltheoretisch modellieren. Ziel dieser Arbeit ist es, aufbauend auf Ergebnisse einers vorherigen Projekts, ein gegebenes spieltheoretisches Modell in ein Protokoll umzusetzen, das jeder Spieler bzw. Teilnehmer befolgt. Hierfür soll die gegebene Simulation erweitert werden. Dieses Projekt wird in Kooperation zwischen den Instituten für Theoretische Informatik und Verteilte Systeme durchgeführt und gemeinsam betreut.

Security assessments of networked systems require knowledge about the utilized communication protocol. For proprietary protocols without known specification and with only limited access to the end-points, the only source of information is the communication itself. To correctly conclude from the captured byte stream to message-formats, -types, and finally a protocol model, structure, message- and field-boundaries, data-type, and semantics need to be inferred.After an initial inference procedure, it is desirable to refine the existing protocol model. Actively probing an entity for the validity of message syntaxes allows to targetedly enhance the knowledge about the protocol. To do this efficiently a smart method of automatically generating test-cases depending on the current protocol model needs to be developed.

Mobile crowdsensing is the method of acquiring user experience data from users. Either in an automated fashion without limited user engagement, for instance using embedded sensors of the smartphone. Or in a participatory fashion, where the user is the main responsible for the provision of data, for instance filling out surveys. With regard to this, we have developed an app that acquires user experience data related to weather in both - automated as well as participatory fashion. However, using multiple embedded sensors of the smartphone consumes resources, battery, as well as storage. For successful completion of the project, the student is required to work on the existing mobile crowdsensing app. More specifically, in the project, the student will have the options to work on optimizing battery consumption, limiting resource usage, optimize sensor data storage, or improve the sensor data accuracy. Or the student has the freedom to suggest his own vision about extending the current app. The successful completion of the project requires the student to actively participate in the project meetings, deliver the tasks on time, write a project report and present their work at the end.

Neurofeedback provides the necessary means to visualize selected and controlled parameters of the brain activity. In healthcare domain, neurofeedback studies enable mitigation of many psychological disorders and illnesses, mainly by therapies that help patients to better self-regulate their brain activity. Electroencephalography (EEG) is the method of monitoring the electrical activity of the brain, thus providing the necessary feedback. In this thesis work, the student is required to survey the current state of frameworks, techniques, or methods that enable coupling of Mobile EEGs with Smartphones. Bluetooth 2.1 with Enhanced Data Rate (EDR) capability is one of the most effective mean of coupling EEGs with Smartphones. The student would therefore be required to work on the Bluetooth stack to acquire real-time data generated from the Mobile EEGs, parse the electrical signal, and visualize the signal semantically. For successful completion of the thesis, the student would be required to identify and address any one of the open challenges faced by the proposed topic. An example of this can be addressing the bandwidth challenges, battery consumption, or signal accuracy

Electroencephalography (EEG) is the method of monitoring the electrical activity of the brain, thus enabling mitigation of many psychological disorders and illnesses, mainly by therapies that help patients to better self-regulate their brain activity. Mobile EEGs are dedicated hardware equipment capable of coupling with many state-of-the-art smartphones. Bluetooth 2.1 with Enhanced Data Rate (EDR) capability is one of the most effective mean of coupling EEGs with smartphones. For successful completion of the project, the student is required to work on Bluetooth 2.1 stack to couple electrical signal simulator with Smartphones. More specifically, in the project, the student will have the options to work on acquiring and collecting data from the simulator, managing the bandwidth of incoming data, real-time data compression, visualizing data on smartphone, or optimally storing data in a database.

As part of our ongoing research, are we currently building a platform for secure statistical analysis based on SGX. The current prototype relies on a very simple statistics language, which we are planning to extend in the future. The goal of this project is to port an existing statistics language interpreter, such as PSPP, to the Rust programming language. Rust features a rich type system and can guarantee memory-safety and thread-safety during compile time, which makes it a great candidate for building safe and fast programming language interpreters. nom is a parser combinators library written in Rust that allows to build safe parsers without compromising on speed or memory consumption. This library can be used as a starting point to implement the parser.

Serverless is a current trend in cloud computing. In contrast to what the name indicates it does not describe an architecture without servers. Instead, it really means that developers do not have to worry about servers and infrastructure, but can completely focus on their code. Unlike previous cloud computing models, a cloud vendor does not offer full platforms or virtual machines, but an execution environment for functions. These often feature a pay-per-use billing model and automatic scalability of resources based on current utilization. Thus, developers are completely relieved of the operational concerns of their applications. All major cloud computing providers offer their own flavor of serverless computing or Function as a Service (FaaS). The goal of this project is to provide a comparison of the similarities and differences of these platforms. Another goal of this project is the implementation of a reference application that can be used to compare different platforms and their programming model. As the practical part of this project a multi node Apache OpenWhisk (an open source serverless platform) cluster should be set up and tested.

Encryption is one of the most reliable techniques for protecting information. However, once data is encrypted, using it becomes very difficult. Goal of this thesis or project, is to explore how Machine Learning algorithms can be designed to be able to deal with encrypted data. Firstly, a survey of existing mechanisms should be conducted. In a second part, algorithms will be comparatively implemented, or own encryption mechanisms introduced.

Mobile crowdsensing is the method of acquiring user experience data from users. Either in an automated fashion without limited user engagement, for instance using embedded sensors of the smartphone. Or in a participatory fashion, where the user is the main responsible for the provision of data, for instance filling out surveys. With regard to this, we have developed an app that acquires user experience data related to weather in both - automated as well as participatory fashion. However, using multiple embedded sensors of the smartphone consumes resources, battery, as well as storage. For successful completion of the project, the student is required to work on the existing mobile crowdsensing app. More specifically, in the project, the student will have the options to work on optimizing battery consumption, limiting resource usage, optimize sensor data storage, or improve the sensor data accuracy. Or the student has the freedom to suggest his own vision about extending the current app. The successful completion of the project requires the student to actively participate in the project meetings, deliver the tasks on time, write a project report and present their work at the end.

Neurofeedback provides the necessary means to visualize selected and controlled parameters of the brain activity. In healthcare domain, neurofeedback studies enable mitigation of many psychological disorders and illnesses, mainly by therapies that help patients to better self-regulate their brain activity. Electroencephalography (EEG) is the method of monitoring the electrical activity of the brain, thus providing the necessary feedback. In this thesis work, the student is required to survey the current state of frameworks, techniques, or methods that enable coupling of Mobile EEGs with Smartphones. Bluetooth 2.1 with Enhanced Data Rate (EDR) capability is one of the most effective mean of coupling EEGs with Smartphones. The student would therefore be required to work on the Bluetooth stack to acquire real-time data generated from the Mobile EEGs, parse the electrical signal, and visualize the signal semantically. For successful completion of the thesis, the student would be required to identify and address any one of the open challenges faced by the proposed topic. An example of this can be addressing the bandwidth challenges, battery consumption, or signal accuracy

Electroencephalography (EEG) is the method of monitoring the electrical activity of the brain, thus enabling mitigation of many psychological disorders and illnesses, mainly by therapies that help patients to better self-regulate their brain activity. Mobile EEGs are dedicated hardware equipment capable of coupling with many state-of-the-art smartphones. Bluetooth 2.1 with Enhanced Data Rate (EDR) capability is one of the most effective mean of coupling EEGs with smartphones. For successful completion of the project, the student is required to work on Bluetooth 2.1 stack to couple electrical signal simulator with Smartphones. More specifically, in the project, the student will have the options to work on acquiring and collecting data from the simulator, managing the bandwidth of incoming data, real-time data compression, visualizing data on smartphone, or optimally storing data in a database.