MLA-C01 Question 170
Select 3You are developing a deep learning model in Amazon SageMaker, and you notice that the model size is significantly larger than expected. Which factors could be contributing to the increased model size?
- A
The number of layers and neurons in the neural network
- B
The amount of training data used during model training
- C
The type of activation functions used in the model
- D
The use of high-dimensional embeddings for categorical features
- E
The precision of the weights (e.g., 32-bit floating-point vs. 16-bit floating-point)
Show answer and explanation
Correct answers: A, D, E
Explanation
The size of a machine learning model is influenced by factors such as the number of parameters (determined by the architecture, including layers and embeddings) and the storage precision of these parameters (e.g., 32-bit vs. 16-bit). However, factors like the amount of training data or the choice of activation functions do not directly affect the model size as they do not alter the number or precision of parameters.
- A. Correct.
The number of layers and neurons directly impacts the model size because more parameters (weights and biases) are needed to connect additional layers and neurons.
- B. Incorrect.
While the amount of training data affects model accuracy and generalization, it does not directly influence the model size, as the model's structure remains the same regardless of the training data size.
- C. Incorrect.
Activation functions do not directly affect the model size. They impact how the model learns but do not contribute to the number of parameters in the model.
- D. Correct.
High-dimensional embeddings for categorical features can significantly increase the model size because embeddings require storage for potentially millions of parameters depending on the vocabulary size and embedding dimension.
- E. Correct.
The precision of the weights affects the size of the model. Using 32-bit floating-point numbers increases the size compared to using 16-bit floating-point numbers, as each parameter requires more storage.