NCA-GENM Question 171
Select 3You are assisting a senior team member in training a multimodal generative AI model. The model's performance is lower than expected, and the senior suggests reviewing the training process for optimization. Which of the following actions should you prioritize to optimize training under their supervision?
- A
Analyze and adjust the learning rate schedule to prevent overfitting or underfitting.
- B
Examine the dataset for class imbalances and apply data augmentation if necessary.
- C
Replace the current optimizer with a random choice of optimizer to see if performance improves.
- D
Increase the batch size without evaluating GPU memory constraints to speed up training.
- E
Monitor and log training metrics like loss and accuracy to identify potential bottlenecks.
Show answer and explanation
Correct answers: A, B, E
Explanation
Optimizing model training involves systematic steps under the supervision of a senior member. Adjusting hyperparameters like the learning rate, ensuring data quality and balance, and monitoring metrics are foundational practices. Random changes or ignoring hardware constraints can lead to inefficiencies or training failures.
- A. Correct.
Adjusting the learning rate schedule is a critical step in training optimization as it directly impacts convergence and model performance.
- B. Correct.
Examining the dataset for class imbalances and applying data augmentation ensures the model learns effectively from diverse and balanced data, which can boost performance.
- C. Incorrect.
Randomly replacing the optimizer without rationale is not a systematic approach to optimization and may lead to unpredictable results.
- D. Incorrect.
Increasing batch size without considering GPU memory constraints is not advisable as it can cause out-of-memory errors and disrupt training.
- E. Correct.
Monitoring and logging training metrics is essential to identify training issues and evaluate the impact of optimization efforts.