Authors
Mohsen Yoosefzadeh-Najafabadi, Valerio Hoyos-Villegas, Ethan Pickering, Seth C Murray, Scott A Jackson
Published in
Trends in plant science. Nov 17, 2026. Epub Nov 17, 2026.
Abstract
The era of big data in agriculture has largely been an era of positive data. Modern breeding preserves cultivars but often loses records of rejected crosses, low-performing lines, and unstable genotype × environment responses. This underutilized dark data constitutes at least 90% of the breeding record, and excluding these data creates survivorship bias that limits AI models trained predominantly on elite material and reduces their ability to learn from rare failures and extreme responses. We propose federated failure learning, a privacy-preserving framework that enables breeding organizations to learn collectively from nonselected material without disclosing proprietary germplasm or raw data. Through homomorphic encryption and federated gradient sharing, isolated institutional observations become a shared predictive shield for global food security.
PMID:
43045546
Bibliographic data and abstract were imported from PubMed on 18 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 23
- Comments 0