Explain overfitting, dropout, normalization, RL post-training

Machine Learning Fundamentals

Answer the following:

1. What is overfitting? What are some ways to mitigate it in machine learning?
2. Specifically for deep learning, what are common techniques used to reduce overfitting?
3. Explain dropout:
   • What does it do during training?
   • Why is it viewed as a form of regularization?
   • How should you handle it at inference time?
4. Compare two common normalization methods used in deep networks (for example, Batch Normalization vs Layer Normalization):
   • Which statistics does each method normalize with?
   • How do their behaviors differ across batch sizes and for sequence models?
   • At deployment, which statistics or parameters are used?
5. Describe common ways reinforcement learning (RL) is applied in LLM post-training (alignment or fine-tuning after pretraining).

Overview: This question evaluates understanding of model generalization and regularization ideas—specifically overfitting, dropout, and normalization techniques—along with the use of reinforcement learning for post-training alignment in large language models.