Harsha Vardhan Tetali

I am a Postdoctoral Researcher in the Department of Computer Science at University of Helsinki, where I work on machine learning models for metabolic systems.

I am passionate about research involving mathematical ideas in machine learning and signal processing. I previously worked on SSD NAND channel modeling at Marvell Semiconductor. My PhD work focused on physics-informed matrix factorizations, with research published in IEEE Transactions on Signal Processing.

Email  /  CV  /  Google Scholar  /  ORCID

My research integrates mathematical concepts from various fields into machine learning, with a focus on physics-informed approaches and signal processing applications. I investigate how domain-specific knowledge can be incorporated into learning algorithms to improve both performance and interpretability.

Incorporating physical constraints and domain knowledge into matrix factorization algorithms improves performance and interpretability for wavefield reconstruction.

Matrix factorization models combined with wave operators enable interpretable wavefield reconstruction for structural health monitoring.

Tensor representations enable improved reconstruction quality for wavefield data in nondestructive evaluation applications.

Theoretical analysis of generalization error in meta-learning using information-theoretic approaches.

Physics-informed machine learning enables accurate dispersion curve estimation in heterogeneous materials.

Unsupervised representation learning with physics-informed constraints enables high-quality wavefield reconstruction.

Dictionary learning with wave physics constraints improves signal representation for guided wave signals.

Dictionary learning techniques enable effective fault detection in photovoltaic systems using spread-spectrum time-domain reflectometry.

Optimal selection of principal components and measurement intervals improves PCA-based anomaly detection in dynamic structural health monitoring environments.

Weight decay optimization enhances unsupervised autoencoder performance for anomaly detection in challenging uncontrolled structural health monitoring scenarios.

Incorporating wave physics into dictionary learning algorithms moves beyond black-box approaches for improved interpretability and performance.

Compressive sensing techniques enable learning of guided wave dispersion curves from complex multi-path reflection patterns.

Theoretical analysis providing global optimality guarantees for wave-informed matrix factorization formulations.

Website design from Jon Barron's template.