MLA-C01 Question 172
Select 2You are training a machine learning model on AWS SageMaker and notice that the model size is significantly larger than expected. Which of the following factors could directly influence the size of your model?
- A
The number of layers and parameters in the model architecture
- B
The data preprocessing techniques applied to the training dataset
- C
The choice of activation functions used in the model
- D
The type of optimization algorithm selected during training
- E
The use of embeddings or large vocabulary sizes in natural language processing (NLP) models
Show answer and explanation
Correct answers: A, E
Explanation
The size of a machine learning model is primarily determined by its architecture, which includes the number of layers, parameters, and specific design choices like embeddings. Factors such as data preprocessing or optimization algorithms influence the training process but do not affect the stored parameters or weights, which are the primary contributors to model size.
- A. Correct.
The number of layers and parameters in the model architecture directly impact the model size because these determine the amount of weights and biases stored during training and inference.
- B. Incorrect.
Data preprocessing techniques do not directly influence model size. These techniques affect the input data but do not alter the architecture or parameters of the model.
- C. Incorrect.
While activation functions impact how the model learns, they do not directly influence the size of the model as they do not affect the stored parameters or weights.
- D. Incorrect.
The type of optimization algorithm influences the training process but does not directly impact the size of the trained model. Model size is determined by the architecture and parameters.
- E. Correct.
Using embeddings or large vocabulary sizes in NLP models can significantly increase the size of the model due to the storage requirements for the embeddings and associated parameters.