This project frames brain age regression as a representation learning problem. The focus is on how data geometry, slicing strategies, and preprocessing assumptions influence learned age-related features, prioritizing interpretability and robustness over architectural complexity.

This work studies unsupervised decomposition of neurophysiological time series, treating PCA and ICA as explicit modeling assumptions about signal generation. Emphasis is placed on component interpretability, cross-subject consistency, and alignment with experimental timing.

This project explores clustering as a representation problem, analyzing how preprocessing pipelines and dimensionality reduction strategies reshape the data manifold and influence downstream interpretability, rather than optimizing for a single clustering score.

This work studies spin correlations in top-quark pair production as a bridge between data-driven analysis and effective field theory, focusing on interpreting statistical observables in terms of SMEFT operators rather than replacing theory with machine learning.

This project treats particle classification as a case study in feature representation and inductive bias, comparing how different algorithms encode physical information under limited supervision and noisy Monte Carlo data.