Universal Spectral Tokenization via Self-Supervised Panchromatic Representation Learning

2025-10-20

AI Summary

Astronomical spectroscopic data exhibit heterogeneity across wavelength domains (optical/infrared), spectral resolution, and astrophysical object types (stars/galaxies), impeding cross-dataset modeling. To address this, we propose the first self-supervised, cross-band and cross-resolution spectral representation model. Our method performs tokenization and representation learning directly on the native wavelength grid—without resampling or pre-alignment—yielding physically interpretable, intrinsically aligned, and homogeneous representations for multi-source spectra. Its key innovation lies in unifying diverse astrophysical objects and spectral domains within a single architecture, while enabling efficient downstream transfer. Experiments demonstrate state-of-the-art performance on stellar parameter estimation and redshift measurement, significantly enhancing cross-dataset information fusion. The framework establishes a generalizable sequence modeling paradigm for astronomical and interdisciplinary scientific foundation models.

Technology Category

• Machine Learning: Representation Learning
• Computer Vision: Representation Learning for Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Sequential scientific data span many resolutions and domains, and unifying them into a common representation is a key step toward developing foundation models for the sciences. Astronomical spectra exemplify this challenge: massive surveys have collected millions of spectra across a wide range of wavelengths and resolutions, yet analyses remain fragmented across spectral domains (e.g., optical vs. infrared) and object types (e.g., stars vs. galaxies), limiting the ability to pool information across datasets. We present a deep learning model that jointly learns from heterogeneous spectra in a self-supervised manner. Our universal spectral tokenizer processes spectra from a variety of object types and resolutions directly on their native wavelength grids, producing intrinsically aligned, homogeneous, and physically meaningful representations that can be efficiently adapted to achieve competitive performance across a range of downstream tasks. For the first time, we demonstrate that a single model can unify spectral data across resolutions and domains, suggesting that our model can serve as a powerful building block for foundation models in astronomy -- and potentially extend to other scientific domains with heterogeneous sequential data, such as climate and healthcare.

Problem

• Unifying heterogeneous astronomical spectra across resolutions
• Overcoming fragmented analysis across spectral domains
• Creating universal representations for diverse scientific data types

Innovation

• Self-supervised learning from heterogeneous spectral data
• Universal tokenizer processing native wavelength grids directly
• Producing aligned homogeneous representations for downstream tasks