MLA-C01 Question 173
Select 2You are building a deep learning model for image classification using Amazon SageMaker. The model is intended for deployment on an edge device with limited memory. Which factors should you consider to reduce the model size without significantly compromising accuracy?
- A
The number of trainable parameters in the model architecture
- B
The size of the dataset used for training the model
- C
The use of techniques like quantization or pruning
- D
The type of activation functions used in the model
- E
The use of pre-trained models with transfer learning
Show answer and explanation
Correct answers: A, C
Explanation
The size of a model is influenced by the number of trainable parameters and optimization techniques like quantization and pruning, which are specifically designed to compress the model. While other factors like dataset size and activation functions influence training and inference efficiency, they do not directly reduce the model's storage footprint.
- A. Correct.
Reducing the number of trainable parameters directly impacts the size of the model as it decreases the memory required to store the weights.
- B. Incorrect.
While the size of the dataset affects training time and model accuracy, it does not directly influence the size of the final model.
- C. Correct.
Quantization and pruning are common techniques to reduce model size by compressing the weights and eliminating unnecessary parameters.
- D. Incorrect.
The type of activation functions primarily influences the computational efficiency during training and inference but has minimal impact on model size.
- E. Incorrect.
Using pre-trained models with transfer learning can improve accuracy with less data but does not inherently reduce the size of the model.