Google Professional Machine Learning Engineer Question 284
Single answerGoogle Cloud PlatformYour team is developing an image classification model using TensorFlow, and you plan to train the model on a large dataset with millions of high-resolution images. The training process is expected to involve complex operations like matrix multiplications and convolutions. You also want to minimize training time while keeping costs manageable. Which compute and accelerator option would be the most appropriate for this use case?
- A
Use a cluster of CPU-based virtual machines for training.
- B
Use a GPU-enabled virtual machine with a high-end NVIDIA GPU.
- C
Use a TPU-enabled virtual machine for training.
- D
Use edge devices with embedded AI accelerators.
Show answer and explanation
Correct answer: C
Explanation
For training a TensorFlow-based image classification model on a large dataset with computationally intensive operations, TPUs are the most appropriate choice. They are optimized for TensorFlow and provide high performance and cost efficiency compared to GPUs and CPUs. While GPUs are also a strong contender, TPUs offer better optimization for the specific workload in this scenario.
- A. Incorrect.
CPUs are not well-suited for computationally intensive tasks like training deep learning models on large datasets. They are slower for operations like matrix multiplications and convolutions compared to GPUs or TPUs.
- B. Incorrect.
GPUs are a strong option for training deep learning models and are widely used for tasks like image classification. However, TPUs are specifically optimized for TensorFlow workloads, making them more cost-effective and efficient for this use case.
- C. Correct.
TPUs (Tensor Processing Units) are designed for high-performance machine learning tasks, especially for TensorFlow models. They are highly efficient for operations like matrix multiplications and convolutions, making them the best choice for this scenario.
- D. Incorrect.
Edge devices with embedded AI accelerators are designed for inference tasks in real-time, low-power environments rather than for training large-scale models. They are not suitable for this use case.