Astrophysics x Graph Machine Learning

Dakshesh Kololgi

PhD researcher at UCL building graph-based machine learning methods to connect galaxies with their dark matter cosmic web environments.

Particle preset

Research Interests

Graph ML for Cosmic Structure

Learning physically meaningful graph representations of galaxy environments using geometry-aware metrics.

Galaxy-Environment Connection

Studying how galaxy properties evolve across voids, walls, filaments, and clusters in the cosmic web.

Simulation to Survey Transfer

Bridging simulation-grounded models toward observational datasets including DESI-era analysis.

Interpretable Scientific AI

Designing ML pipelines that remain transparent, testable, and anchored in astrophysical understanding.

Latest Paper

Learning the Cosmic Web: Graph-based Classification of Simulated Galaxies by their Dark Matter Environments

We introduce a three-stage simulation framework: T-Web environment labels, graph features from Delaunay triangulation, and a graph attention network (GAT) classifier.

  • Task: classify each galaxy as void, wall, filament, or cluster from local structure.
  • Result: 85% accuracy for galaxies with stellar mass above 10^9 M_sun.
  • Impact: graph-based representations improve environment inference and support future DESI-focused studies.

View on arXiv