← Back to projects

Research Project

When Is Uncertainty Informative?

An independent research project studying when uncertainty helps an active learner choose informative observations, and when model assumptions or distorted representations make uncertainty misleading. Gaussian mixture models provide a controlled laboratory for the investigation.

  1. 01Implemented and tested an expectation-maximization (EM) Gaussian mixture model on controlled synthetic data
  2. 02Compared entropy-based querying with random selection across 100 fresh trials, including replication checks and statistical tests
  3. 03Investigated initialization sensitivity, class imbalance, outliers, model misspecification, and alternative uncertainty estimates
  4. 04Compared Gibbs sampling with mean-field variational inference for posterior uncertainty over component means

Python · NumPy · scikit-learn · Jupyter · Git

When Is Uncertainty Informative?, screenshot 1 of 3
When Is Uncertainty Informative?, screenshot 2 of 3
When Is Uncertainty Informative?, screenshot 3 of 3