Active Spectral Learning

Project overview

Bayesian machine learning project on active learning for Gaussian Processes with spectral kernels. The project studied how active sampling can improve learning of non-stationary spectral structure.

Motivation

Efficiently learn spectral structure with limited labels and uncertainty-aware query selection.

Method

Gaussian Processes with spectral kernels and active learning criteria driven by predictive uncertainty.

Evaluation

Compared uncertainty-aware sampling against passive baselines on reconstruction and generalization metrics.

Links

PDF (placeholder URL)