publications

2026

1. In prep.

   

   Interpretable Machine Learning for Stellar Spectral Classification with LAMOST DR5 and Gaia DR3

   Alex Baker

   2026

   Manuscript in preparation

   Abs Code Website

   This work develops an interpretable machine-learning pipeline for stellar spectral classification using physics-based spectroscopic descriptors extracted from LAMOST DR5 spectra cross-matched with Gaia DR3. The pipeline combines feature engineering, supervised classification, dimensionality reduction, and SHAP-based interpretability to study which spectral features drive model decisions. Current results suggest that metallicity-sensitive indicators, including Ca II H&K and Mg b, play a major role in the learned classification structure, alongside classical temperature-sensitive Balmer features.

2. Course

   

   Réduction de dimension et interprétabilité appliquées aux spectres stellaires LAMOST DR5

   Alex Baker, Justine Jean, and Nérimantas Caillat

   2026

   Course report and symposium contribution, PHY-3500, Université Laval, Winter 2026. Supervised by Prof. Antoine Allard.

   Abs Code

   Course research project presented in the context of PHY-3500 at Université Laval. The project applies dimensionality-reduction methods and interpretability tools to LAMOST DR5 stellar spectra in order to compare PCA, UMAP, t-SNE, and autoencoder-based representations of astrophysical feature space. The work builds on the AstroSpectro pipeline and focuses on the physical interpretation of low-dimensional spectral structure.

3. Draft

   

   Deep Visual Representations for Galaxy Morphological Classification

   Alex Baker

   2026

   Project manuscript draft in preparation

   Abs Code

   This project explores deep-learning methods for galaxy morphological classification using Galaxy10 DECaLS, DINOv2 visual features, non-parametric morphometrics, photometric information, and uncertainty-aware classification. The work investigates whether self-supervised visual representations encode physically meaningful galaxy structure beyond class labels alone.

2025

1. Software

   

   \(ξ\) Dark Energy · BAO MCMC: Exploratory Bayesian Inference for Dark-Energy Models

   Alex Baker

   2025

   Exploratory research-software release, v0.1.0

   Abs Code Website

   This project implements a reproducible Bayesian MCMC workflow for comparing ΛCDM, CPL, and exploratory oscillating dark-energy models using Pantheon+SH0ES supernovae and DESI DR2 BAO measurements. The release focuses on transparency, posterior diagnostics, model comparison, and explicit documentation of current limitations rather than claiming evidence for new physics.