Câu 14: RAI: Risk and Artificial Intelligence

A risk analyst is designing a credit default model using data that is partially unlabeled. To address the issue of overfitting, the analyst is considering a co-training approach. Which of the following would be a reason for choosing co-training to mitigate overfitting?

1. A. Co-training uses dimensionality reduction and only keeps relevant features.
2. B. Co-training utilizes different views of feature inputs to improve generalization. — đáp án hiện tại
3. C. Co-training uses ensemble methods to average predictions from different models.
4. D. Co-training is effective against overfitting even when its underlying assumptions are violated.