The University of Stavanger invites applications for a doctorate scholarship in Computer Science, Signal Processing, or Cybernetics at Department of Electrical Engineering and Computer Science.
This is a trainee position that will mainly give promising researchers an opportunity for professional development leading to a doctoral degree.
The research fellow will be appointed for three years with only research duties or four years with research and 25% compulsory duties. This will be clarified in the recruitment process. The position is vacant from 01.07.2017 or by appointment.
It is possible to apply on up to three of the following projects: 1. Developing a new algorithm for real-time process mining based on Petri Nets 2. The control systems behind biomolecular sensing and their connection to systems for adaptation 3. New Applications of Smart Contracts and Block Chain Technology 4. A low-cost, power efficient audio amplifier using digital technology („all-digital class-D amplifier”) 5. Spatio-temporal analysis of transportation data 6. Image segmentation for medical applications 7. Deep Learning from Sparsely Labeled Data 8. Data-driven approach to analysis and prediction of heart rate using pulse watches
Please mark in your application order of preference.
For further information regarding required qualifications and each project, please see below.
The position is funded by Norwegian Ministry of Education and Research.
Applicants must have a strong academic background with a five-year master degree, preferably recently, or possess corresponding qualifications which could provide a basis for successfully completing a doctorate. Both the grade for the master’s thesis and the weighted average grade of the master’s degree must individually be equivalent to or better than a B grade.
By rating it will be placed on the applicant’s potential for research in the field, as well as that person’s individual prerequisites for research education.
The appointee must be able to work independently and as a member of a team, be creative and innovative. The research fellow must have a good command of both oral and written English.
This fellowship position is important for obtaining a scientific position at a University.
The doctorate will mainly be carried out at the University of Stavanger, apart from a period of study abroad at a recognized and relevant centre of research.
The research fellow is salaried according to the State Salary Code, l.pl 17.515, code 1017, LR 20, ltr. 50 of NOK 435 500 per annum.
The position provides for automatic membership in the Norwegian Public Service Pension Fund, which guarantees favourable retirement benefits. Members may also apply for home investment loans at favourable interest rates.
Project description and further information about the position can be obtained from Head of Department Tom Ryen, telephone +47 51832029, email: firstname.lastname@example.org
Information about the appointment procedures can be obtained from HR consultant Anne Karin Rafos, telephone +47 51831711, email email@example.com
The University is committed to a policy of equal opportunity in its employment practices. The University currently employs few female research fellows within this academic field and women are therefore particularly encouraged to apply.
Please register your application in an electronic form on jobbnorge.no. Relevant education and experience must be registered on the form. Certificates/diplomas, references, the list of publications and other documentation that you consider relevant, should be submitted as attachments to the application as separate files. If the attachments exceed 15 MB altogether, they will have to be compressed before uploading.
1. Developing a new algorithm for real-time process mining based on Petri Nets Powered by the fields like machine learning and artificial intelligence, Process Mining is a new type of Big Data Analytics for optimizing production processes. Process Mining is to detect the digital footprint that is left behind in the IT systems by the production processes. There are algorithms that are under use for process mining, to understand what happened during the production processes and to propose improvements to the processes. In this project, we shall develop a new algorithm that is based on Petri nets, a discrete mathematical tool for modeling, simulation, and performance analysis of discrete event systems. In this research, we shall study about process mining, and then develop an efficient Petri nets based algorithm that can autonomously mine the production data real-time. Supervisors: Professor Reggie Davidrajuh, firstname.lastname@example.org and dr.ing. Nejm Saadallah, IRIS
2. The control systems behind biomolecular sensing and their connection to systems for adaptation A hallmark property of living systems, from single cells to complex organisms, is the ability to sense the outside world. Many biochemical sensors can be said to function as differentiators. They react to changes in the outside environment, often with large signal responses that return to basal levels if the change is sustained. This type of behavior is closely related to adaptation and homeostatic mechanisms. An adapting sensor that gives a response and then returns to its original pre-stimuli state, when faced with sustained stimuli, can function over a larger dynamic range in stimuli input. Applying control theoretic methods and nonlinear analysis, one goal of this project is to identify ways biochemical species in an organism can interact to form simple structures that work as sensors. Another goal is to examine the similarity and connection between adapting systems and sensors in biological systems. A large portion of the work will be theoretical, but the right candidate will also have the possibility to study and implement systems in simple cells, such as bacteria or yeast, in the laboratory.
The applicant should have a master’s degree with specialization in control theory, dynamical systems, or other related fields. A background in biology is not required, but certainly a plus, necessary courses in biology will be offered.
The successful applicant will join an interdisciplinary group with researchers and other PhD students from both control engineering and molecular biology. Supervisors: Associate professor Kristian Thorsen, email@example.com and professor Tormod Drengstig, firstname.lastname@example.org
3. New Applications of Smart Contracts and Block Chain Technology Bitcoin represents a revolution in online payment between untrusted participants. The Bitcoin payment system is made possible by a sophisticated combination of cryptographic mechanisms and a distributed transaction processing mechanisms backed by a globally distributed transaction ledger. This is often referred to as block chain technology. Block chains guarantees that transactions recorded in the chain cannot be tampered with. These block chains can facilitate a wide range of novel applications, that have been impossible to achieve with per-block chain technology.
The goal of the project is to develop new and innovative applications powered by block chain technology and smart contracts. One such novel application domain is document verification with embedded biometric templates, where the existence of the document and its authenticity can be verified using a public block chain.
A successful applicant should have a master’s degree in computer science with focus on distributed systems and/or computer security. Supervisors: Professor Hein Meling (UiS), email@example.com and professor Roman Vitenberg (UiO).
4. A low-cost, power efficient audio amplifier using digital technology („all-digital class-D amplifier”) In an all-digital power amplifier, each audio sample is mapped into a fixed amplitude pulse where the width is proportional to the sample height, this is known as pulse-width-modulation (PWM). That is, the digital signal from a CD, or other sources, is converted to a pulse train with digitally controlled pulse widths. The pulse train is then sent to a transistor switching stage (H-bridge) for amplification, followed by an LC filter combined with a loudspeaker. The advantages of this scheme compared to traditional, analog amplifiers are: 1) Power efficiency: More than 90% of the power is converted to sound waves, compared to a theoretical 25% for class-A, and 78.5% for class AB amplifiers. 2) The signal chain is kept digital all the way to the loudspeaker/LC filter, thus avoiding noise and distortion along the way. Two main problems with all-digital class-D amplifiers are: 1) The mapping problem: The mapping of height (signal amplitude) into pulse width gives rise to nonlinear distortion. 2) The dead-time problem: The transistor output stage cannot switch instantly. There is a dead-time, around 10-20 ns, when both transistors are closed. This dead time leads to incorrect pulse widths and nonlinear distortion results.
The project will focus on theoretical as well as practical problems in solving the dead-time problem. One possible method is to compute (in real time) the inductor current at each switching point in time. With this knowledge *beforehand*, the dead time problem can be avoided by appropriate adjustment of the switching time instant. This method can be combined with a practical reduction of the problem by using new transistor technology to develop switching stages with shorter dead-time. An additional goal of this project is to build a complete class-D amplifier with solutions for both the mapping problem and the dead time problem, and to quantify its performance.
The candidate should have a master degree with a solid background in signal processing, as well as in measurement methods and practical electronics (microcontrollers, digital signal processors, and analog design). Supervisors: Professor Sven Ole Aase, firstname.lastname@example.org and associate professor Kristian Thorsen, email@example.com
5. Spatio-temporal analysis of transportation data Spatial (geographical) indexing structures such as the R-tree have long been used in traditional database systems to speed up geographical queries. However, there has been much less work on using such data structures in big data systems such as Hadoop or Spark. A significant amount of spatial data also has a temporal dimension. A good example is bus and traffic monitoring data. The task in this project is to study how to apply spatial and temporal indexing structures for transportation data analysis using the data analytics system here at UiS, possibly in connection with Stavanger Smart City efforts as part of the Triangulum project. In big data systems such as Hadoop or Spark, the idea is to split the work between the nodes in a distributed computer with a shared file system but no shared memory. Therefore, a spatiotemporal query will require two things: 1) A strategy for filtering the data so that the nodes only need to process data that are relevant to the query. This filtering step is what spatial and spatiotemporal indexing structures have performed for traditional database systems. 2) A strategy for dividing the data set between the nodes in such a fashion that the nodes have the data they need for processing. Thus, each node can work on its own portion of the data set with as little need as possible for communication with other nodes.
One subtask in this project is to study existing spatial and temporal indexing structures and determine which would be most suitable for our tasks. This involves: 1) Look at existing spatial and spatiotemporal indexing structures to determine which are best suited to a distributed systems. 2) Determine whether the indexing structures can also be used for dividing the data between nodes. Indexes already split the data into chunks based on some criterion (different for different indexes), so this might be possible. 3) Different queries seem to require different splitting and filtering strategies. Can a single index be created that is applicable to a wide variety of queries, or do the indexes need to be built for particular queries? 4) On our data set some queries to look at time in a linear fashion (example: find bus delays during Tour des Fjords). Other queries look at time in a repeating periodic fashion (example: find bus delays on Sundays). How does this affect the indexing schemes? Most spatiotemporal indexes are built with the first time view in mind. Consider the representation of temporal data both in temporal and frequency domain w.r.t storage, indexing, and analysis.
Another subtask would be to look at existing systems such as SpatialHadoop and OpenTSDB and see what they can do and whether they are applicable to our problem. The final subtask would be to implement a spatio-temporal indexing structure on the UiS cluster (based on Spark) and test it with the data mentioned earlier, possibly collaborating with master students or the PhD student on the “Improving Public Infrastructure through Data Science” project.
Supervisors: Associate professor Erlend Tøssebro, firstname.lastname@example.org and Tomasz Wiktorski, email@example.com
6. Image segmentation for medical applications There is a wide range of needs for image processing, segmentation and classification within medical applications with different image modalities ranging from X-ray, CT, MRI and photographic images. The biomedical data analysis group at UiS has for many years collaborated with different departments at the local hospital, as well as with other national and international partners. The need for segmentation has been obvious in many ongoing collaborative projects. In this proposed project the focus is on the segmentation methodologies where we will look at X-ray, CT and MR images of the heart as example of segmentation needs in addition to other example images. The project aims at tailoring classical segmentation algorithms for specific applications, but also in developing new ideas within segmentation of (medical) images. Level sets and Graph cuts are examples of state-of-the-art methods in image segmentation that will be explored in the cardiac image applications. Deep learning neural networks have provided state-of-the-art results in a number of computer vision tasks in later years. Recent work has utilized deep learning networks for medical image segmentation, thus this will be given considerable attention as well.
The applicant should have a Master’s degree with specialization in signal- and image processing, and/or pattern recognition/machine learning. Supervisor: Professor Kjersti Engan, e-mail: Kjersti.Engan@uis.no
7. Deep Learning from Sparsely Labeled Data Deep neural networks, a.k.a. deep learning, have transformed the fields of computer vision, speech recognition and machine translation, and now rivals human-level performance in a range of tasks. While the idea of neural networks dates several decades back, their recent success is attributed to three key factors: (1) vast computational power, (2) algorithmic advances, and (3) the availability of massive amounts of training data. There is no doubt that deep learning will continue to transform other fields as well, including that of information retrieval. One major challenge is that for most information retrieval tasks, training data is not available in huge quantities. This is unlike, for example, to object recognition, where there are large scale resources at one’s disposal to train neural networks with (tens of) millions of parameters (e.g., the ImageNet database contains over 14million images).
Deep learning is inspired by how the brain works. Yet, humans can learn and generalize from a very small number of examples. (A child, for example, does not need to see thousands of instances of cats, in many different sizes and from numerous different angles, to be able to recognize a cat and tell it apart from a dog.) Can deep neural networks be enhanced with this capability, i.e., to be able to learn and generalize from sparsely labeled data? The aim of this project is to answer this question, specifically, in the application domain of information retrieval.
The candidate is required to have a background in machine learning or information retrieval. Supervisor: Professor Krisztian Balog, firstname.lastname@example.org
8. Data-driven approach to analysis and prediction of heart rate using pulse watches Pulse watches have become a popular training aid for many sports, in particular cycling and running. However, the quality of data and usability for diagnostics and predicting future performance remain unknown. HR watches produce a significant amount of data, what calls for automated data analysis and requires the application of big data tools in addition to a mix of other data science concepts such as machine learning and time series analysis. The research will be performed primarily based on data obtained during NEEDED study, containing HR, ECG, blood samples, and other data for over a 100 subjects collected during “Nordsjørittet” race. In addition to HR data from pulse watches, other types data can also be used to improve the quality of analysis, these include: population data, ECG, blood samples, etc. While they open for more in - depth analysis, they also increase the complexity of algorithms. The goal is to develop a model to predict HR of a subject during various parts of the race based on earlier performance and other available data.
The data abundance might be useful in early stages of analysis. Nevertheless, it is important to determine a minimal set of inputs that can provide useful and conclusive information to the user and physician. Another important question is to what extent causal conclusion could be drawn large observational data and if it is possible to design simple controlled experiments to strengthen the causal argument. These questions will drive further stages of the PhD work.
Students with both computer science, data science, biomedical engineering, and signal processing background are welcome to apply. Knowledge of Python will be an advantage. Supervisors: Associate professor Tomasz Wiktorski, Tomasz.email@example.com, professor Trygve Eftestøl, Trygve.firstname.lastname@example.org and professor II Stein Ørn, email@example.com