Google Professional Machine Learning Engineer
Get started today
Ultimate access to all questions.
Ultimate access to all questions.
You are working as a machine learning engineer for a cybersecurity organization and your task is to develop a system log anomaly detection model. After developing the model using TensorFlow, the next step is to use it for real-time predictions. You plan to create a Dataflow pipeline to ingest data via Pub/Sub and write the results to BigQuery. Considering factors such as deployment, serving latency, and efficiency, what is the best approach to minimize the serving latency while deploying your TensorFlow model for real-time prediction?
Explanation:
The best approach to minimize the serving latency while deploying your TensorFlow model for real-time prediction is to deploy the model to a Vertex AI endpoint and invoke this endpoint in the Dataflow job. Vertex AI provides a fully managed, serverless platform optimized for real-time inference, allowing for high performance and low latency. By leveraging Vertex AI endpoints, Dataflow can easily invoke the model for predictions without the overhead of loading the model as a dependency, which would add unnecessary latency and complexity. Additionally, Vertex AI offers built-in tools for model monitoring, TensorBoard, and model registry governance, making it the most efficient and scalable option for real-time prediction.