Selected by the College of Doctoral Studies, this recognition highlights the exceptional quality and impact of Johan Larsson's research. The award, which includes a monetary prize of SEK 25,000 from the LUSEM Partnership, will be officially presented and handed over in June.
Congratulations, Johan! Can you briefly describe your thesis and its main findings?
”My thesis is primarily about making it faster to fit statistical models to high-dimensional data, where there are many more variables than there are observations. A typical example of this is genetic data, where a data set may contain millions or even billions of genetic markers (variables) but only a few hundred individuals (observations). I have worked on models that help identify which of these variables are important in any given data set. Fitting statistical models in this setting is extremely costly computationally and the work that I have presented in the thesis helps to reduce this burden, sometimes cutting the time you have to wait for your computer to finish fitting a model from hours to minutes.”
What inspired you to choose this topic?
”I was fascinated by a talk that Malgorzata Bogdan, my then-future supervisor, gave at the department at the start of my PhD. She talked about SLOPE, which is one of the statistical models that my thesis revolves around. I learnt that there were several open problems with respect to the computational side of SLOPE and quickly became involved in trying to solve those.”
How does your research contribute to the field or have practical implications?
”The main implication of my work is that some of these models, like SLOPE, have become accessible to a wider audience. It was previously the case that researchers would turn to other, simpler, models for the practical reason that they did not have the computational resources required to tackle large-scale data, but can now make do with a regular laptop.”
What does winning this award mean to you?
”I am grateful and honored to receive this award and the encouragement it brings me. I would like to thank the school and all of my previous colleagues and co-authors, especially my main supervisor Jonas Wallin who was supportive and encouraging throughout all of my PhD.”
Thank you, Johan! Wishing you all the best in your future research and career.
The nomination
Below is part of the nomination text for Johan's dissertation award, which ultimately led to Johan receiving the prize.
In his dissertation, Johan Larsson develops new algorithms and optimization methods for sparse regression—a methodology used when the number of predictors is large relative to the number of observations. Special emphasis is placed on regularization methods, which introduce a penalty on model complexity to reduce the risk of overfitting. This is crucial in statistics and machine learning, where models with many variables often risk capturing random noise rather than meaningful signals. Regularization helps identify the most important variables, much like finding a needle in a haystack, making theoretical models more applicable to real-world data. The dissertation introduces new algorithms for sparse regression, where many predictors can be excluded without losing essential relationships, with a particular focus on regularization methods such as LASSO and SLOPE.
Other parts of the nomination highlighted that all contributions in the dissertation are of a high technical level and directly applicable in statistics and machine learning for analyzing complex and high-dimensional data. It also stated that several key results have been published at leading conferences such as NeurIPS and AISTATS and that during the Ph.D. period, Johan developed a new course in data visualization for statisticians, which was highly appreciated. The doctoral work is also described as being characterized by a high degree of independence. The nomination was written by Johan’s supervisor Jonas Wallin, Senior lecturer at Department of Statistics.