Contact: Simin Nadjm-Tehrani
Mixed reality is an umbrella term for any combination of reality with virtual elements. However, the algorithms used typically use video processing and rendering techniques that require a lot of computation and their complexity is increasing. This makes it hard (or even impossible) to run them in good conditions in a user (mobile) device for longer periods due to resource constraints.
Edge computing is a recent paradigm that aims at providing resources accessible with lower latency and increased privacy compared to cloud resources. By enabling the offloading of the computation-heavy task to a more powerful edge device while keeping the latency low, edge computing promises to enable applications like mixed reality.
In an earlier work, a mixed reality prototype using edge computing was implemented, its performance evaluated, and the resulting workload characterized. The first part of the application is a client running on a smartphone and responsible for getting the video input and (after treatment at the server) displaying the mixed reality video output to the user. The second part of the application runs on the server (at the edge) and is responsible for doing the heavy computations needed for providing the mixed reality content. A few enhancements to this prototype regarding hardware acceleration and encoding have been implemented. The current prototype uses WiFi for the client/server communication.
The aim of this thesis is to upgrade the application with regards to newer libraries to realise the function and also perform experimental evaluations over a 5G network, thereby understanding the performance gains and bottlenecks over 5G.
Contact: Simin Nadjm-Tehrani
Edge computing is a recent paradigm that aims at providing distributed resources accessible closer to the user compared to cloud resources. These resources can then handle a variety of tasks locally, which reduces the amount of data being sent in the network, and also enables getting context-aware data.
One of the challenges of edge computing is that the workload to be handled is very dynamic and varying in time and space. Therefore, it is important to be able to monitor for local load spikes in order to adapt the resources allocated to maintain a good quality of service. What would be even more interesting is to be able to predict in which areas the load spike will happen in order to decrease the time where the service has to perform in a degraded way due to overload.
In a previous work, we proposed to use mobile edge devices to handle these local load spikes. The spikes where predicted thanks to a threshold mechanism, which serves as input for deciding where to dynamically allocate the mobile resources. The aim of the thesis would be to investigate and implement a learning-based solution to predict the forthcoming local load spikes.
The thesis project will be carried out utilizing national resources at the EdgeLab ( AI-Sweden).
Contact: Simin Nadjm-Tehrani
Supervisory control and data acquisition (SCADA) systems control and monitor modern critical infrastructure such as electricity distribution systems. Given the emerging of attacks targeting the control system of critical infrastructures, SCADA security has become a pressing issue.
Network-based anomaly detection is one of the potential defence mechanisms for SCADA security. It aims to model the normal SCADA traffic, e.g. using machine learning techniques, and issue an alarm when there are deviations from the learnt model. Previous work models the IEC-60870-5-104 (IEC-104) traffic from a real electric utility with its timing attributes and the model performs well in presence of non-trivial attacks.
This thesis project aims to extend the timing model to electric utilities that run different protocols such as Generic Object Oriented Substation Event (GOOSE) and Manufacturing Message Specification (MMS). Using the characteristics of the acquired data sets, may lead to discovery of new models or alternative learning approaches.
Contact: Simin Nadjm-Tehrani
The time spent in verification for safety-critical software is an order of magnitude larger than the time spent in actual development of the critical functions. Such software systems are at the core of avionics and autonomous/semi-autonomous vehicles, drones, traffic-signaling systems, and railway systems. Any means to increase the speed of development without losing control of the safety case will be vitally important with increased reliance on software in cyber-physical and embedded systems. In earlier research (DOI: 10.1109/HASE.2008.59) a method is proposed for analyzing exposure of systems, built by composing several components, to single or double faults at the interface of some component. The components are mathematically modelled, and those faults that the system is provably tolerant to are already dealt with (checked by deterministic model checking). This paper tackles the faults for which one cannot show whether the system is tolerant of, or not. This happens where exhaustive formal verification is unfeasible, hence a probabilistic approach is used. The aim of this thesis project is to build upon this work and to evaluate the industrial application of this modelling framework with respect to technical feasibility. The focus will be modelling a safety-critical, cyber-physical software system and evaluating probabilities for catastrophic events at the Critical Systems unit at Combitech. You will decide, together with your supervisor, the precise software system to model early in the thesis work.
Contact: Simin Nadjm-Tehrani
En grupp energibolag i Sverige har avsikten att kika på hur förbrukningsbeteenden skiljer sig mellan deras respektive områden. Var för sig har de tex uppgifter kring folks elförbrukning, värmeförbrukning, bredbandsuppkoppling, vattenförbrukning, geografi mm. För att fördjupa dessa insikter kring kunder baserat på denna data men även för att utforska potentialen med att kombinera datan med extern data tänker de skapa en förståelse och se vilka mönster som skulle kunna hittas ifall de tex kombinerade vår data med data från fastighetsregistret, väderdata, inkomster, fordonsdata, antal familjemedlemmar. Exjobbet går ut på att:
Contact: Simin Nadjm-Tehrani
Recent advances in machine learning are now being applied in safety-critical systems where software defects may cause severe harm to humans and the environment. Providing convincing arguments that such software-based systems are safe and correct is problematic due to human´s inability to understand the software, and thus fails to identify scenarios in which the software should be tested in. Several researchers have proposed formal verification methods that address these concerns, but the scalability of these methods put limitations on the size and complexity of the software being analyzed.
This thesis work aims to assess trade-offs between different machine learning models when the ability to argue for safety and correctness is important. A prototype of an airborne collision avoidance system is provided, implemented as a neural network, with formal requirements that must be verified for such systems to be deployed in the real world. The work will be done in collaboration with the Avionics Platform & Services division at Saab Aeronautics.
Contact: Simin Nadjm-Tehrani
Ett företag som levererar olika samhällskritiska tjänster har flera olika delsystem som är mer eller mindre beroende av varandra. Dessutom upphandlas produkter och tjänster från tredje part. System som kan ha otydliga beroenden kan vara för komplexa för att ha en helhetssyn vad gäller kritikalitet och arkitektur. Målet med detta examensarbete är att med hjälp av tekniska redskap kartlägga systemkomponenter som är nödvändiga för att leverera en organisationstjänster, deras beroenden som uppfattas av olika aktörer inom organisationen, och hur incidenthantering för olika delsystem sköts. Utgångspunkten ska vara leverans av tjänsten och dess tillgänglighet. Kartläggningen sker genom informationssamling inne i organisationen eller underleverantörer (med betoning på molntjänster) och leder till ett underlag som kan användas för att skapa en generell metod för att effektivt göra likartade kartläggningar hos andra organisationer. Detta exjobb görs i samarbete med Svenska Kraftnät (Svk) inom ramen av forskningscentret RICS (www.rics.se).
Contact: Simin Nadjm-Tehrani
Att säkerställa informationssäkerhet hos ett nätverkat system, t.ex. kritiska infrastrukturer som levererar samhällskritiska tjänster, bygger på analys av systemet under olika förutsättningar inklusive eventuella attack scenarier och felyttringar. För att kunna utföra systematiska och repeterbara tester där olika försvarsmekanismer utvärderas behöver man utsätta samma normalscenario för en mängd olika attacker och detektioner. Experiment med olika utgångspunkter kan skapas i en test nätverk där flera aktörer (forskare, leverantörer) kan testa sina prototyper och produkter. Därmed behöver data vara frikopplad från en viss organisation och inte kunna spåras till de ursprungliga miljöer där det skapades i. Målet med detta exjobb är att skapa syntetisk data som liknar realistisk data genom att "tvätta" data som normalt inte skulle vara tillgänglig inom en organisation, men som kan delas med andra aktörer under ordnade former. Exjobbsarbetet ska resultera i en metod med tillhörande verktyg som skapar syntetiska datan samt definiera kriterier och metrik för att validera både "likheten" med det ursprungliga datan och dessa anonimitetsegenskaper. Detta exjobb görs i samarbete med forskningscentret RICS (www.rics.se) och kommer till nytta för dess avnämare.