Tutorial
2026 workshop program
Introduction to Transformers
The work
Abstract
This hands-on tutorial explores the application of transformer-based deep learning to real astronomical data from the ALeRCE broker. We will query some real data and train two transformer models from scratch: a sequence transformer for classifying sources from multi-band light curves, and a Vision Transformer (ViT) for classifying objects from image stamps. The session closes with a short demo of how query capabilities can be wrapped as an AI agent tool, illustrating how classical astronomy pipelines integrate into modern agentic workflows. Participants should have basic Python familiarity; PyTorch experience is helpful but not required.
Requirements: GitHub Repo