liu-johandahlin

Johan Dahlin Passionate about finding hidden patterns and trends in data.

  • 2016

    PhD Automatic Control

    Linköping University, Sweden.
    Includes two years of coursework in Engineering, Machine Learning and Statistics.

  • 2011

    MSc Engineering Physics

    Umeå University, Sweden.
    Majors in Risk Management and Industrial Statistics.

  • 2011

    B.Sc. Economics

    Umeå University, Sweden.
    Includes coursework in Marketing, Project Management and Law.

Biography

I began life in Enköping, Sweden in 1986. As a young boy, I quickly found an interest in Mathematics and Computer Science, which led me to study Engineering and then to pursue a PhD based on research and coursework. 

I received my PhD in Automatic Control in May 2016 after successfully defending my thesis which contained a total of 17 peer-reviewed papers (13 conference papers and 3 journal papers) published at top conferences and journals in the fields of Computational Statistics and System Identification.

Currently, I am developing automated data-driven algorithms for building dynamical models as a PostDoc at the School of Engineering at the University of Newcastle, Australia.

During my research career, I have worked at a number of different companies and universities. I visited Prof. Robert Kohn at he University of New South Wales,  Australia during the autumn of 2014 as part of his PhD studies. I have also worked as a Research Scientist at Sectra AB and as a PostDoc at the division of Statistics and Machine Learning at Linköping University.

My objectives

The Economist has described data as the new oil which can result in the next industrial revolution. Everyday, more and more information is gathered from sensors and the Internet. As a result, Big Data, Statistics and Machine Learning (in short Data Analytics) have become essential tools to generate insights from this data.

The objective in these fields is often to condense the data into a model, which can be used to gain understanding, to make decisions or to forecast future behaviours. However, many models and the algorithms that fit them to data cannot yet cope with the complex, dirty and large data materials. More work is required to better scale with the amount of information and to accelerate current methods for model building.

I am passionate about finding hidden patterns and trends in data such as numbers, images and documents to learn more about medicine, economics and society at large. During my PhD studies, I spent five years learning and developing more efficient algorithms for Data Analytics.

I now aim to make use of my expertise to improve the world in terms of health care, medicine and the environment. I believe that learning from data is essential in this quest by enabling improved data-driven decision making and forecasting.

Skills

I have acquired quite a few different skills and a lot of knowledge during my more than one decade at different universities. Some are soft skills like problem solving, communication, teaching, managing projects and people and handling stress. Some are more traditional skills like Mathematics, Physics, Computer Science, Statistics, Economics, etc. Most relevant for my work today is my skills in Statistics, Machine learning and Programming, which I have broken down below.

Statistics

Bayesian modelling
Time series analysis
Monte Carlo methods
Multivariate data analysis
Inference

Machine learning

Linear regression and classification
Deep learning with images
Gaussian processes
Reinforcement learning
Topic modelling of texts

Programming

Python
R
C(++)
JS, Redux, React
GIT

Research interests

My interests in research are rather broad and spans employing methods from Statistics, Machine Learning, Artificial Intelligence, Engineering, etc. to solve real-world problems. You can see some examples of practical applications of my research on the next page and here I discuss some more academic interests.

Accelerating Monte Carlo algorithms for Bayesian inference, such as particle filtering and Markov chain Monte Carlo.

Efficient inference in longitudinal data, i.e., short data records for many individuals.

Approximate Bayesian inference by modifying the model to simplify inference, e.g., using Gaussian process optimisation.

Developing methods for applying Machine/Deep Learning for applications in Climate Science, Medicine and Finance.

Applications of my research

This page contains some examples of existing and possible applications of my research. I have worked on most of these projects myself as part of my PhD studies and my PostDocs. Click on the title to learn more about each application.

Connect with me

I love to connect with new people for networking and to share experiences and idea. Therefore, do not hesitate to get in touch with me if you have any questions connected to my research, the source code on GitHub or if you would like to discuss a business idea or hire me for a job.