Speaker
Description
Current exoplanet detection relies heavily on iterative sampling and box-searching, methods that face significant scaling challenges with the massive data volumes expected from next-generation surveys. This presentation introduces an exploratory concept: adapting Large Language Model (LLM) architectures to treat photometric light curves as "textual" sequences. By utilizing self-attention mechanisms, we propose a framework to identify planetary transits and regress orbital parameters (Rp/Rs, period) in a single forward pass. This approach aims to leverage the "global context" of transformers to better distinguish subtle planetary signals from stellar noise and instrumental systematics. While this work is in its conceptual infancy, we outline a roadmap for using these models as rapid inference engines to provide high-quality priors for traditional physical modeling, potentially accelerating the discovery of small planets in big-data archives.