Natalí de Santi

Natalí de Santi

PhD in Physics

Predicting galaxy bias using machine learning

Published in arxiv, 2026

Catalina Riveros-Jara et al 2026 arXiv https://arxiv.org/abs/2602.05881

Paper highlights:

  • Galaxy bias can be predicted per galaxy: using data from IllustrisTNG300, we show that the individual linear bias parameter $b_i$ can be modeled directly from halo and environmental properties, moving beyond population-averaged bias.
  • Environment is the dominant driver: local overdensity (especially $\delta_8$) is the most informative feature, followed by distance to cosmic-web structures and halo formation time. Bias is primarily an environmental quantity.
  • Probabilistic ML is essential: normalizing flows outperform deterministic models because they capture the intrinsic stochasticity of the matter-halo-galaxy connection, recovering not just the mean bias, but its full distribution.

You can read the paper here.