method · first seen 7 Nov 2026
black-box model evaluation without held-out test set
Also: evaluability
Evaluating a model's properties when test samples may have leaked into training, using only black-box access to outputs.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Toward a theory of EvaluabilityPapers · 7 Nov 2026 · Kunal Talwar
Latest sources
- Toward a theory of EvaluabilityPapers · 7 Nov 2026 · Kunal Talwar