Project LiU1: Analysis and resolution of conflicts in dependable autonomous systems

Project LiU2: Bayesian Learning for Spatio-Temporal Processes

Project LiU3: Probabilistic-Logic Stream Reasoning and Learning for Safe Autonomous Systems

Project LiU4: Identity-Defined Networking: Securing the Industrial Internet

Project LiU5: Structure based learning for autonomous systems

Project LiU6: Size Efficient Direction of Arrival Estimation

Project LiU7: Automatic pattern tracking and visualization for multi-field data

Links To Application Portal:

Application deadline: June 15 2017

Project LiU1: Analysis and resolution of conflicts in dependable autonomous systems

Supervisor: Simin Nadjm-Tehrani, Dept. of Computer & Information Science, Linköping University

Contact info: Phone +46 (0)13 282411, Mail: simin.nadjm-tehrani@liu.se

Short Project description: Conflict resolution for autonomous systems will gain more attention with real deployments in safety-critical systems. Conflicts can arise within a system, e.g. where predefined rules for enhancing safety may jeopardize security of a system (leading to vulnerabilities that can be exploited by an adversary). Conflicts can also arise among multiple actors in a cooperative scenario where the complexity in the goal-action-state space for multiple agents may lead to unintended, resource-wise suboptimal, or unsafe states. Conflicts arising from different actions can be analyzed at two stages: (1) design-time analysis aims to identify and avoid conflicting actions, e.g. those that a safety analysis process and a security analysis process may identify, each plausible in their own context but not consistent, (2) run-time conflicts that will need to be dealt with in presence of a dynamic external context of high complexity. The latter will be similar to detecting real-time anomaly detection and mitigation. This project will explore a combination of graph-theoretic, game-theoretic, or model-based analysis techniques for the design space resolution, and machine learning for the dynamic case.

Project LiU2: Bayesian Learning for Spatio-Temporal Processes

Supervisor: Mattias Villani, Division of Statistics and Machine Learning, Dept. of Computer and Information Science, Linköping University.

Contact info: Phone +46 070-0895205, Mail: mattias.villani@liu.se

Short Project description: The rapid deployment of streaming sensors have made spatio-temporal data increasingly common. In addition, devices ranging from small sensors to fully autonomous cars are more and more interconnected in networks and hierarchies. The project will develop probabilistic models for spatio-temporal data with associated computationally efficient Bayesian learning methods. Emphasis will be given to network data. Potential modeling frameworks could be multi-output Gaussian processes and Gaussian random fields. The aim is to exploit the commonalities of many different application areas with spatio-temporal data to develop general models and learning methods. The methodological research will be applied to several different application areas, for example problems in transportation research, such as real-time urban road traffic prediction and resilience of transportation networks. The candidate is expected to obtain a PhD in Computer Science or Statistics.

Project LiU3: Probabilistic-Logic Stream Reasoning and Learning for Safe Autonomous Systems

Supervisor: Fredrik Heintz, Department of Computer Science, Linköping University

Contact info: Phone +46 700 895689, Mail: fredrik.heintz@liu.se

Short Project description: One of the major open problems in artificial intelligence is how to combine the power of first-order logic and probability theory to represent, learn and reason with complex and uncertain relational structures in a principled manner. The purpose of this project is to develop incremental techniques for learning and reasoning with such statistical-relational models. The inference has to be incremental to handle the velocity and volume of information produced by autonomous systems. To adapt to new situations the reasoning should be continually improving through semi-supervised on-line learning. The developed methods and techniques will be applied to make autonomous systems more safe through continuous and context-dependent monitoring.

Project LiU4: Identity-Defined Networking: Securing the Industrial Internet

Supervisor: Andrei Gurtov, Department of Computer Science, Linköping University

Contact info: Phone +46 700850566, Mail: andrei.gurtov@liu.se

Short Project description: The current Internet networking is based on TCP/IP protocol stack that had not changed significantly for 40 years. If the future Internet-of-things, smart cities and Industrial Internet would use the same model, the collapse is imminent due to widespread cybersecurity risks. The last year, we witnessed 1-Tbps Denial-of-Service attacks from hacked IoT devices, enough to take a small country out of the Internet. A recent scanning study in Sweden revealed thousands of sensitive industrial devices connected and open for attacks. Securing current networks using firewalls, segmentation and Virtual Private Networks (VPNs) is complex, costly and fragile. It requires plenty of manual configuration which is not sustainable in the long run. The root defect is the use of ephemeral identities such as IP and link addresses to define the policies. We propose to develop a novel approach based on cryptographic host identities and Host Identity Protocol (HIP), a new standard by the Internet Engineering Task Force (IETF). With help of centralized orchestration, it reduces network provisioning time, decreases costs, and reduces the attack surface. In this project, we plan to study scalability, resilience and performance of IDN for the Industrial Internet.

Project LiU5: Structure based learning for autonomous systems

Supervisor: Erik Frisk, Dept. of Electrical Engineering, Linköping University

Contact info: +46 13 28 5714, Mail: erik.frisk@liu.se

Short project description: Autonomous operation of dynamic systems with high uptime requirements and no direct user feedback is a challenging task where efficient data analytics and learning is crucial. Direct data analytics, without considering the time varying behavior, may lead to reduced inference accuracy. The key vision here is to utilize so called structural models in machine learning. A structural model is a graph, a description without equations that can handle much larger models than analytical equation based approaches. A core feature of integrating the structure graph in learning is increased accuracy, managing concept drift, reduced need for data imputation, and more efficient utilization of existing data. Results will have applications in operation, fault management, flexible maintenance, and autonomy.

Project LiU6: Size Efficient Direction of Arrival Estimation

Supervisor: Gustaf Hendeby, Dept. of Electrical Engineering, Linköping University

Contact info: Phone +46 13 28 58 15, Mail: hendeby@isy.liu.se

Short Project description: Triangulation using direction of arrival (DoA) to available signals is one of the fundamental methods for positioning.  The project will explore DOA estimation by using the local phase of the signal rather than its estimated second order moments as in conventional methods in literature, with the potential to downscale the size of the array one order of magnitude.  The results would provide the basis for more efficient positioning systems, suitable for e.g., internet of things (IoT), as an alternative in indoor and GPS-denied environments.

Project LiU7: Automatic pattern tracking and visualization for multi-field data

Supervisor: Ingrid Hotz, Dept. of Science and Technology, Linköping University

Contact info: Phone +46 11 363462, Mail: ingrid.hotz@liu.se

Short Project description: Automatic tracking and highlighting of visually identified patterns plays an important role in monitoring of autonomous systems. Thereby patterns of interest can be of different type and are often based on multiple data sources. They can exhibit complex shapes and structures. Translating such patterns into mathematically tractable properties it is often a challenging task. The goal of this work is to establish methods to track patterns defined by selection in visualizations of multi-field data. Expressive descriptors and similarity measures will be developed which are detailed enough to encode the relevant information about the pattern but are also general enough to allow variations in terms of size, deformation and orientation. The descriptors will be based on moments, which are robust, flexible and can serve as an excellent tool to construct invariants that do not change under certain transformations.  Based on the extracted tracks of patterns a visualization system will be developed highlighting their development and changes.