What’s New? / About Me / Publications / Music

---

About Me

I’m a researcher at Nokia Bell Labs, Espoo, Finland. I received the D.Sc. (Tech.) degree in signal processing technology from Aalto University, Espoo, Finland, in 2022. My doctoral thesis developed theory and methods for high-dimensional covariance matrix estimation (more info below). I hold both a B.Sc. (Tech.) and a M.Sc. (Tech.) in electrical engineering from Aalto University, Espoo, Finland, as well as a Bachelor of Culture and Arts degree in pop and jazz music from Metropolia University of Applied Sciences, Helsinki, Finland. My interests include jazz music, statistical signal processing, machine learning, and wireless communications.

You can find more about my research from the links provided in the Publications page. Many of my articles also include code, which can be downloaded from my GitHub page.

Doctoral research

My doctoral research centered on theoretical and applied research in the intersection of signal processing, multivariate statistics, and machine learning. I worked in Prof. Esa Ollila’s research group with the Department of Signal Processing and Acoustics at Aalto University, Finland. The focus of my research was on advanced learning methods and optimized algorithms for estimation and classification. Specifically, I developed theory and methods for high-dimensional covariance matrix estimation for low sample size problems.

In statistical data analysis and machine learning problems, it is important to be able to measure and quantify the degree of associations between the variables. This information is given by the covariance matrix, which is a building block for many important techniques in statistics and machine learning, such as in classification and clustering algorithms. Covariance matrix estimation methods developed for the traditional data regime often perform poorly in high-dimensional low sample size settings. A common approach to solve the problem is to use some variable selection or dimension reduction technique.

In my doctoral research, I have developed regularized covariance matrix estimation methods for high-dimensional single class and multiclass problems. Regularization refers to incorporating additional or prior information into the problem, which can be expressed in the form of structure in the unknown parameters, which can entail sparsity, low rank, or the parameter space can be constrained to lie on a manifold. Using these tools and techniques, we have been able to derive various computationally efficient and accurate estimation methods, whose usefulness has been demonstrated using real world data in problems such as classification or portfolio optimization in finance.