Google Professional Machine Learning Engineer Question 357
Select 3Google Cloud PlatformYou are deploying a machine learning model to predict customer churn for a subscription service. The model must process predictions in under 50 milliseconds per request, handle up to 10,000 concurrent requests, and minimize memory usage due to limited resources in the production environment. Which of the following actions should you take to optimize the ML solution for these requirements?
- A
Use TensorFlow's model optimization toolkit to apply quantization.
- B
Switch the deployment environment to a GPU-accelerated machine.
- C
Serve the model using Google Cloud's Vertex AI Prediction with autoscaling enabled.
- D
Reduce the size of the input features by selecting only the most important ones.
- E
Increase the batch size for prediction requests to maximize GPU utilization.
Show answer and explanation
Correct answers: A, C, D
Explanation
To optimize the ML solution for low latency, high throughput, and minimal memory usage, applying quantization reduces the model's size and speeds up inference. Vertex AI Prediction with autoscaling ensures the system can handle concurrent requests efficiently. Feature selection reduces memory and computational overhead by limiting unnecessary input data. Switching to a GPU or increasing batch size may improve throughput but can introduce latency, making them less ideal for the specified requirements.
- A. Correct.
Using TensorFlow's model optimization toolkit to apply quantization reduces the size of the model and improves inference speed, making it suitable for low-latency and low-memory environments.
- B. Incorrect.
Switching to a GPU-accelerated machine may increase performance but can also increase latency for single predictions due to batch processing overhead, making it less ideal for real-time, low-latency requirements.
- C. Correct.
Using Vertex AI Prediction with autoscaling ensures the system can handle up to 10,000 concurrent requests by dynamically scaling resources according to demand.
- D. Correct.
Reducing the size of the input features by selecting only the most important ones decreases the memory footprint and processing time, improving overall performance.
- E. Incorrect.
Increasing the batch size is useful for maximizing GPU utilization in batch processing but can introduce latency, making it unsuitable for scenarios requiring low-latency predictions.