The validation phase of the AI life cycle focuses on tuning the model and selecting the best architecture before final evaluation. Statement II, adjusting hyperparameters like learning rates using a held-out validation dataset, is a core activity of this phase. Statement III, selecting the best model type (e.g., CNN vs. RNN) based on performance on the validation set, is also a key part of this phase. Statement I, reporting final accuracy on the test set, occurs in the subsequent Testing/Evaluation phase. The test set must remain unseen during training and validation. Statement IV, using backpropagation, is the fundamental process of the Training phase, where the model learns from the training data.

A. This option is incorrect because backpropagation (IV) is part of the Training phase, not Validation.

B. This option is incorrect because executing the model on the Test Set (I) is part of the final Testing/Evaluation phase, not Validation.

C. This option is incorrect because statement I (using the Test set) and IV (backpropagation) are not part of the Validation phase.

1. Ng, A. (n.d.). Machine Learning Yearning. Stanford University. Chapter 5, "Basic Recipe," clearly distinguishes between the training set (for learning parameters), the development (validation) set (for tuning hyperparameters and model selection), and the test set (for final, unbiased performance evaluation).

2. Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep Learning. MIT Press. Section 5.3, "Hyperparameters and Validation Sets," p. 119, explains that the validation set is used to "tune the hyperparameters" and that the test set is used only at the very end to evaluate final performance.

3. National Institute of Standards and Technology (NIST). (2023). AI Risk Management Framework (AI RMF 1.0). Section 4.3, "Test, Evaluation, Validation, and Verification," describes validation as confirming that a system meets its intended purpose, which includes the iterative tuning process.