I am an Electrical Engineer with a PhD from Uppsala University. I received my Master of Science degree in Electrical Engineering from Uppsala University in 2018. My main research interests are signal processing and machine learning with overparameterized models, and right now I’m looking for opportunities where I can apply my deep knowledge in the fields of machine learning and signal processing to create product value using machine learning and AI solutions.

---

Presentations

Presentation, September 15th 2022, at RISE Research Institutes of Sweden, “Fake it until you make it: Fake features can improve estimation performance.”, in their Seminar Series “Learning Machines Seminars”, where I present our work on Fake Features.

---

List of Publications

Distributed Continual Learning with CoCoA in High-dimensional Linear Regression, 2023, Preprint, currently under revision</a>

Authors: Martin Hellkvist, Ayca Özcelikkale, Anders Ahlén.

We characterize the generalization error obtained when using the distributed learning framework CoCoA to perform continual learning.

Estimation under Model Misspecification with Fake Features, 2023, Published in IEEE Transactions on Signal Processing

Authors: Martin Hellkvist, Ayca Özcelikkale, Anders Ahlén.

We characterize the estimation error in a statistical learning setting with both fake and missing features.

Regularization with Fake Features, 2022, Presented and published via EUSIPCO 2023

Authors: Martin Hellkvist, Ayca Özcelikkale, Anders Ahlén.

We analytically and numerically investigate the trade-offs between the presence of fake features and the explicit ridge regularization.

Continual Learning with Distributed Optimization: Does COCOA Forget?, 2022 (Preprint)

Authors: Martin Hellkvist, Ayca Özcelikkale, Anders Ahlén.

We explore continual learning from the persepective of distributed learning.

Model Mismatch Trade-offs in LMMSE Estimation, 2021, Presented and published via EUSIPCO 2021

Authors: Martin Hellkvist, Ayca Özcelikkale.

We characterize how the performance of LMMSE estimation is affected by missing features in the model.

Linear Regression with Distributed Learning: A Generalization Error Perspective, 2021, Published in IEEE Transactions on Signal Processing

Authors: Martin Hellkvist, Ayca Özcelikkale, Anders Ahlén.

We give bounds on the generalization error in a distributed learning setting for a large family of regressor distributions.

Generalization Error for Linear Regression under Distributed Learning, 2020, Presented and published via SPAWC 2020

Authors: Martin Hellkvist, Ayca Özcelikkale, Anders Ahlén.

We consider a distributed linear regression problem, and characterize the genrealization error for Gaussian regressors.

---

Teaching Experience

1TE661: Signals and Systems, 2018

Extent: BSc. course, approx. 100 students.

Duties: Lecturing, flipped classroom tutorials, oral examinations and grading of written exams.

1TE651: Signal Processing, 2019

Extent: MSc. course, approx. 50 students.

Duties: Tutorial classes.

1TE689: Graphical Programming in LabVIEW, 2020, 2021, 2022

Extent: BSc. course, approx 50 students.

Duties: Tutorial classes and oral examination. Remote and on-campus experience.

1TE717: Digital Technology and Electronics, 2019, 2021, 2022

Extent: BSc. course, approx. 90 students.

Duties: Tutorial classes, responsible of organizing and supervising electronics projects, grading of project reports, grading of orally presented hand-in assignments. Remote and on-campus experience.