New “nearest neighbor” algorithms may lift the curse of dimensionality

The “Painted Ladies” houses in San Francisco can illustrate the nearest neighbor method in computer and systems sciences. Photo: Ross Joyner/Unsplash.

On March 5, 2024, Sampath Deegalla successfully defended his PhD thesis at the Department of Computer and Systems Sciences (DSV). He has focused on the so-called nearest neighbor method and developed new algorithms. We’ve asked him about his research.

“My thesis examines the effectiveness of the nearest neighbor method in high-dimensional datasets such as microarrays, chemoinformatics and images, focusing on overcoming the curse of dimensionality through dimensionality reduction,” says Sampath Deegalla.

He has developed several nearest neighbor algorithms which employ dimensionality reduction for improved accuracy.

A selfie in the DSV kitchen after the thesis defence. Photo: Sampath Deegalla.

“The research explored the impact of feature fusion and classifier fusion techniques and the formation of nearest neighbor ensembles using dimensionality reduction. Empirical studies demonstrated that feature fusion and classifier fusion could significantly enhance nearest neighbor algorithm accuracy, with the choice of dimensionality reduction method playing a critical role,” he continues.

Deegala explains that the findings can be used by different stakeholders:

“Researchers and data scientists who are dealing with high-dimensional data, such as in image processing, genetics, or any field where data dimensionality poses a challenge, could benefit from my findings. Policymakers and organizational leaders might also find value in my work by understanding how to use machine learning techniques to make effective decisions,” says Sampath Deegalla.

I have contributed to the body of knowledge with practical implications

He currently works at the University of Peradeniya in Sri Lanka and finishing the PhD thesis is of course an important milestone.

“Completing my PhD at DSV, Stockholm University is a significant achievement in my academic and professional journey. It reflects my contributions towards advancing the field of data science and machine learning. I have contributed to the body of knowledge with practical implications for handling high-dimensional data, which is increasingly prevalent in various domains.”

“I wish to continue as a senior lecturer at the Faculty of Engineering, University of Peradeniya, focusing on further research, teaching, and mentoring students in data science and machine learning. The findings of this thesis could be used in my teaching and new research projects,” Deegalla concludes.

Sampath’s efforts have borne fruit

From the left: Henrik Boström, Sampath Deegalla and DSV’s Sindri Magnússon at the defence. Photo: Private.

When Sampath Deegalla started out on his journey as a PhD student at DSV, Professor Henrik Boström became his supervisor. In 2017, Henrik Boström started to work at KTH but kept his supervising duties at DSV. He is also happy to see Dr. Sampath Deegalla’s work in print.

“One of the qualities of Sampath that stands out the most is his endurance. Despite a significant work load over many years as a teacher in his home country, he never gave up the idea of completing his PhD studies,” says Henrik Boström.

“Now, almost two decades after starting, Sampath’s efforts have borne fruit in a successfully defended PhD thesis. His scientific contributions have indeed shown to stand the test of time, with some of his early papers still receiving large numbers of citations. As a supervisor, I am very happy and proud of his achievement,” Boström says.

Text: Åse Karlén