{"paper":{"title":"Optimizing Deep Learning Photometric Redshifts for the Roman Space Telescope with HST/CANDELS","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A new semi-supervised model PITA outperforms other methods for photometric redshifts by training on both labeled redshifts and all available images and colors.","cross_cats":["astro-ph.GA"],"primary_cat":"astro-ph.IM","authors_text":"Ashod Khederlarian, Biprateep Dey, Brett H. Andrews, Jeffrey A. Newman, Tianqing Zhang","submitted_at":"2026-02-10T19:01:05Z","abstract_excerpt":"Photometric redshifts (photo-$z$'s) will be crucial for studies of galaxy evolution, large-scale structure, and transients with the Nancy Grace Roman Space Telescope. Deep learning methods leverage pixel-level information from ground-based images to achieve the best photo-$z$'s for low-redshift galaxies, but their efficacy at higher redshifts with deep, space-based imaging remains largely untested. We used Hubble Space Telescope CANDELS optical and near-infrared imaging to evaluate fully-supervised, self-supervised, and semi-supervised deep learning photo-$z$ algorithms out to $z\\sim3$. Compar"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our new semi-supervised model, PITA (Photo-z Inference with a Triple-task Algorithm), outperformed all others by learning from unlabeled and labeled data through a three-part loss function that incorporates images and colors for all objects as well as redshifts when available.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That performance gains measured on HST/CANDELS imaging will generalize to Roman Space Telescope data characteristics and that latent space smoothness directly improves photo-z accuracy without overfitting or domain shift issues.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"PITA, a new semi-supervised deep learning algorithm, outperforms prior photo-z methods by using a triple-task loss on images, colors, and available redshifts to produce a smooth latent space.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A new semi-supervised model PITA outperforms other methods for photometric redshifts by training on both labeled redshifts and all available images and colors.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"25d0024ec1ab175003f001e965d02190b7e059d9773d0f19db2c915c1624ccd5"},"source":{"id":"2602.10207","kind":"arxiv","version":2},"verdict":{"id":"332a442e-3432-4ed9-911d-e7062d606aba","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T02:21:02.372234Z","strongest_claim":"Our new semi-supervised model, PITA (Photo-z Inference with a Triple-task Algorithm), outperformed all others by learning from unlabeled and labeled data through a three-part loss function that incorporates images and colors for all objects as well as redshifts when available.","one_line_summary":"PITA, a new semi-supervised deep learning algorithm, outperforms prior photo-z methods by using a triple-task loss on images, colors, and available redshifts to produce a smooth latent space.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That performance gains measured on HST/CANDELS imaging will generalize to Roman Space Telescope data characteristics and that latent space smoothness directly improves photo-z accuracy without overfitting or domain shift issues.","pith_extraction_headline":"A new semi-supervised model PITA outperforms other methods for photometric redshifts by training on both labeled redshifts and all available images and colors."},"references":{"count":226,"sample":[{"doi":"10.48550/arxiv.2101.04293","year":2021,"title":"2021, arXiv e-prints, arXiv:2101.04293, doi: 10.48550/arXiv.2101.04293","work_id":"d8627c77-420f-4836-8614-6f0bc45abd5e","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.3847/1538-4365/ab929e","year":2020,"title":"A., Almeida , A., et al","work_id":"ccf01265-1391-4620-ba31-7803e222071e","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1051/0004-6361/202347395","year":2024,"title":"2024, A&A, 683, A26, doi: 10.1051/0004-6361/202347395","work_id":"9a4db10c-4895-4138-a940-83163fca942b","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.3847/2041-8213/adab76","year":2041,"title":"Akins, H. B., Casey, C. M., Berg, D. A., et al. 2025a, ApJL, 980, L29, doi: 10.3847/2041-8213/adab76","work_id":"9c8376af-e403-4271-858e-524caa1f260b","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.3847/1538-4357/ade984","year":null,"title":"Akins, H. B., Casey, C. M., Lambrides, E., et al. 2025b, ApJ, 991, 37, doi: 10.3847/1538-4357/ade984","work_id":"998294a1-dc8b-4cca-9c19-c1fe64467656","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":226,"snapshot_sha256":"034fd08631b37d0054b1fde10e175428a07048b7acf21c206eb479fbd3722b93","internal_anchors":13},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}