Google Professional Machine Learning Engineer Question 286
Single answerGoogle Cloud PlatformYou are designing a machine learning pipeline to train a deep learning model for image classification on a dataset containing millions of high-resolution images. The model uses convolutional neural networks (CNNs) and requires significant parallel processing power for both training and inference. Cost efficiency and scalability are key considerations. Which compute and accelerator option would be the most appropriate for this scenario?
- A
Use CPUs on Google Compute Engine for both training and inference.
- B
Use GPUs on Google Compute Engine for training and CPUs for inference.
- C
Use TPUs on Google Cloud for training and inference.
- D
Use GPUs on Google Compute Engine for both training and inference.
Show answer and explanation
Correct answer: C
Explanation
TPUs are the most appropriate choice for this scenario due to their specialized design for deep learning workloads. They offer high parallelism and scalability, which are critical for training CNNs on large datasets of high-resolution images. Additionally, TPUs are cost-efficient for both training and inference in Google Cloud, making them a better option compared to CPUs or GPUs in this case.
- A. Incorrect.
CPUs are not well-suited for the computational demands of training deep learning models on large datasets. They lack the parallel processing power required for efficient training, making this option unsuitable.
- B. Incorrect.
While GPUs are a good choice for training deep learning models, using CPUs for inference may not be efficient in this case, as the model requires significant computational resources even during inference. This option does not fully address scalability and performance needs.
- C. Correct.
TPUs (Tensor Processing Units) are specifically designed for high-performance machine learning tasks, particularly for deep learning models like CNNs. They provide exceptional parallel processing capabilities and are cost-effective for large-scale training and inference in Google Cloud.
- D. Incorrect.
Using GPUs for both training and inference can be a viable option, but TPUs typically offer better performance and cost efficiency for deep learning tasks at scale, particularly for frameworks like TensorFlow.