Abstract

This thesis demonstrates the successful application of transfer learning to bridge ground-based and space-based spectroscopic analysis through adapting SpecPT (Spectroscopy Pre-trained Transformer) for Hubble Space Telescope WFC3 grism data for redshift prediction. Originally trained on high-resolution DESI spectra, SpecPT initially failed when applied directly to low-resolution, noisy HST WFC3 grism observations (Normalized Median Absolute Deviation (NMAD) = 0.2095, catastrophic outlier fraction ($\eta$) = 47.97\%). Through transfer learning on 8,530 high-quality 3D-HST spectra with emission-line SNR > 2.5 and z < 1.7, the model achieved substantial improvement (NMAD = 0.0724, $\eta$ = 26.69\%), representing a 65\% reduction in typical redshift error and 46\% decrease in catastrophic failures. The research addresses two primary objectives: establishing transfer learning effectiveness for cross-domain spectroscopic analysis and investigating whether supplementing grism spectra with broadband photometric data enhances performance. Counterintuitively, integrating comprehensive multi-wavelength photometric data from CANDELS significantly degraded performance (NMAD = 0.1641, $\eta$ = 36.51\%), challenging conventional astronomical assumptions about multi-modal data fusion and revealing critical failure modes in astronomical machine learning. This work establishes a unified framework for automated analysis of both ground-based and space-based spectroscopic surveys, with important implications for JWST, Euclid, and the Nancy Grace Roman Space Telescope. The demonstrated capability to adapt models across instrumental domains provides a scalable approach for processing large data volumes from next-generation missions, validating foundational model approaches that can be developed once and efficiently adapted across diverse observational contexts.

Publication Date

4-2026

Document Type

Thesis

Student Type

Graduate

Degree Name

Astrophysical Sciences and Technology (MS)

Department, Program, or Center

Physics and Astronomy, School of

College

College of Science

Advisor

Jeyhan Kartaltepe

Advisor/Committee Member

Andrew Robinson

Campus

RIT – Main Campus

Download

Share

COinS