Validation Machine Learning
noun
Definition 1: The use of a held-out data, or cross-validation, to evaluate a model during development and to support model selection or hyperparameter tuning [Machine Learning Glossary].
Example in context: “The use of cross-validation over holdout validation is particularly advantageous with health care data sets that are often comparatively small to moderately sized, costly to obtain, or restricted by privacy and regulatory concerns.” [Wilimitis et al. 2023]
Definition 2: The verification of the correctness, admissibility, or consistency of data, inputs, or system architecture independently of model-quality evaluation [ISO/IEC 22989:2022].
Example in context: “Based on these two observations, we propose two new input validation methods based on local robustness verification, which can protect neural networks from adversarial examples, especially from strong attacks, and improve their accuracies on clean data.” [Liu et al. 2020]
Synonym: model validation (sense 1); input checking (sense 2)
Related terms: validation set, cross-validation, hyperparameter tuning, early stopping, input validation, data validation, robustness