Swedish Study Group: Mathematics in Society
August 19-24, 2018

KTH Royal Institute of Technology
Stockholm




Projects


Project descriptions:


Data driven modeling for control of industrial systems
The main goal of the study is to implement control algorithms for a special type of propulsion systems using machine learning. The problem itself is rather complex and can be solved in several steps. The first step is to create data based models for inverse systems that will then be used to implement control solutions. Data based modeling becomes more and more important due to the increasing trends of Big Data and Artificial Intelligence/ Machine Learning. Technically, this problem belongs to the area of numerical analysis and neural networks modeling. The task is to create a neural network model based on given data. Data are presented by input matrix and output vectors. This neural network should be used to fit the output in the best way. The second part of the problem is to create another neural network model fitting the input by the output in the best way, an "inverse" model in some sense. This part requires some extra analysis since the function mapping the input into output is not one-to-one function but a surjective function, meaning that different input values can be mapped to the same output. Thus the question about the adequacy of the inverse model is quite important. Numerical properties of both models and their performances should be carefully analyzed.

Detailed description

Team Leader: Xiaoming Hu, Dept. Mathematics, KTH


Ahum – the first AI-supported matching service for psychological therapy
Traditionally, the psychotherapist's skills and experience have been seen as globally suited for all types of psychotherapeutic needs and the primary mechanism for client behavioral changes in dynamic psychotherapy. Conversely, the therapist's personality has been virtually ignored or regarded as inconsequential to therapeutic outcome. Therapists are, however, human, so when encountering clients, they bring moral and cultural attitudes in spite of efforts to maintain objectivity and neutrality. Hence, the personality of the therapist influences the psychotherapeutic process. Ignoring the impact of the above factors may lead to sub-optimal therapy for the client, so there is a need for an objective evidence based scheme for matching clients with the most appropriate therapist. The goal of this project is to use machine learning in order to suggest the most suitable options to a patient based on their data input and taking also into account the personality and skills of the therapists.

Team Leader: Pawel Herman, EECS, KTH


Mathematical simulation and parameter identification of a bolometer
A microbolometer is a device for measuring the power of incident electromagnetic radiation via the heating of a material with a temperature-dependent electrical resistance in infrared (IR) cameras. In the camera, an array of bolometers gives an IR image. At Flir in Täby, we develop and produce cameras for temperature measurement. Each pixel of the IR image is represented by a readout voltage, which is a noisy signal. This voltage is the signal that is measurable at the moment. However, in the literature, there are mathematical models describing the underlying behaviour of the bolometer, before the readout voltage. We would like to investigate if the underlying bolometer parameters can be identified using readout voltage data. It is also interesting to get an understanding of how the properties of the parameter estimates are affected by noise. The long term goal is to get better understanding of the components (detectors) that we purchase and to strengthen our capabilities to reduce noise.

Team Leader: Olof Runborg, Dept. Mathematics, KTH