AstroAI Lunch Talks - February 25, 2026 - Nolan Koblischke
25 Feb 2026 - Joshua Wing
The video can be found here: https://www.youtube.com/watch?v=yY1cCh4P3DE
Speaker: Nolan Koblischke (University of Toronto)
Title: AION-Search: Semantic search for 100M+ galaxy images using AI-generated captions
Abstract: Finding scientifically interesting phenomena in billions of galaxy images currently relies on slow manual labeling campaigns. We build a semantic search engine from completely unlabeled image data by leveraging Vision-Language Models (VLMs) to generate descriptions for galaxy images, then contrastively align a pre-trained astronomy foundation model with these embedded descriptions. We find that current VLMs generate descriptions that are sufficiently informative to train a semantic search model that has strong zero-shot performance on rare phenomena despite no deliberate curation for rare cases during training. We further introduce a VLM-based re-ranking method that nearly doubles recall for our most challenging targets. AION-Search enables flexible semantic search for 100 million galaxy images, and our approach generalizes to making large unlabeled scientific image archives semantically searchable across domains.