Yu-Yang-Li/StarWhisper

What it solves

StarWhisper is a comprehensive suite of AI tools designed for astronomy, addressing the challenges of analyzing vast amounts of astronomical data and automating the operation of telescopes. It provides specialized models for interpreting light curves, spectra, and textual research, as well as frameworks for autonomous observation planning.

How it works

The project is organized as a collection of specialized AI modules:

• Astronomical LLMs: Models trained on astronomical Q&A, research texts, and code to assist in scientific inquiry.
• Time-Series Analysis: Tools for classifying variable stars and pulsars using light curves (including sparse data from ZTF/ATLAS).
• Spectral Analysis: Models designed to analyze stellar spectra, specifically those with low signal-to-noise ratios.
• Telescope Automation: A framework for end-to-end autonomous astronomical observations, including planning, execution, and anomaly handling.
• Observation Support: Tools for generating masks from all-sky camera images and estimating supernova explosion times.

Who it’s for

This project is intended for astronomers, astrophysicists, and AI researchers focusing on the same domain who need tools for automated data classification, autonomous telescope control, and AI-assisted research writing and analysis.

Highlights

• End-to-End Automation: Includes a framework for automating the entire observation process from planning to execution.
• Domain-Specific LLMs: Provides weights for StarWhisper 3, a model tailored for astronomical research.
• Linguistic Approach to Spectra: Treats stellar spectra as a language for analysis.
• Multi-Source Data Support: Handles data from various sources like Kepler, K2, ZTF, and ATLAS.