Secure Collaborative ML Prototyping with Agent Platform Workbench
Short answer
D Create Agent Platform Workbench instances with GPUs and preinstalled PyTorch, leveraging managed identity for secure BigQuery access without manual key management.
Question
• It must provide access to GPU accelerators.
• It must allow seamless authentication to BigQuery without manual management of service account keys.
• It must be pre-configured with common ML frameworks.
You need to choose the most secure solution with the least setup overhead for your team to begin model development and experimentation. What should you do?
Key Takeaways
The main trap is choosing a solution that offers technical capabilities but fails on security (manual keys) or workflow fit (batch training vs. interactive prototyping). Managed identity is the critical differentiator for secure, low-overhead access.
This question tests selecting the optimal tool for secure, collaborative ML prototyping. The correct solution leverages Agent Platform Workbench for managed GPUs and pre-configured frameworks while using managed identities for secure BigQuery access.
Common mistake: Option B is a common distractor because it mentions GPU-enabled serverless training. However, serverless training is for executing jobs, not for the interactive, collaborative exploration required in the initial prototyping phase described in the scenario.
Expert Analysis
The scenario requires a collaborative, cloud-based prototyping environment with GPU access, pre-configured ML frameworks (PyTorch), and secure BigQuery authentication without manual key management. The key constraints are security (avoiding service account keys) and minimal setup overhead.
Managed Identity vs. Service Account Keys: Option D uses the instance's managed identity to access BigQuery. This is the most secure approach because credentials are automatically rotated and never exposed to users or stored in files. Options A and C explicitly mention downloading or managing service account keys, which violates the security requirement and increases administrative overhead.
Collaborative Prototyping vs. Batch Training: Option B suggests using Agent Platform Serverless Training. While this supports GPUs, it is designed for executing training jobs, not for interactive, collaborative prototyping where engineers need to iteratively explore data and code in notebooks. Option A involves setting up a JupyterHub cluster on Compute Engine, which requires significant manual configuration of drivers and libraries, contradicting the "least setup overhead" requirement.
Workbench Capabilities: Option D proposes creating Agent Platform Workbench instances. These are managed notebook environments that support GPU attachment and come with pre-installed kernels for popular frameworks like PyTorch. They integrate seamlessly with Google Cloud services, allowing direct access to BigQuery via the client library using the attached service account (managed identity). This meets all criteria: GPU access, pre-configured frameworks, secure authentication, and low setup effort.
Community comments confirm that Workbench provides the necessary managed infrastructure and secure identity integration, whereas other options either lack collaboration features, require excessive manual setup, or compromise security.
Exam Strategy
When asked for a prototyping environment, prioritize managed notebook solutions (like Workbench or Colab Enterprise) over self-managed VMs or batch training services. Always check for 'managed identity' or 'workload identity' as the preferred method for accessing other GCP services securely, avoiding any option that requires manually handling service account JSON keys.
Frequently Asked Questions
Why is managed identity better than service account keys?
Can I use Colab Enterprise instead of Workbench?
Community Votes
100% of anonymous learners picked answer D. Votes are pick records left by other test-takers — they are not the verified answer.
Discussion 2 contributions
Compiled from public community threads and reader submissions, lightly edited for clarity. Handles and vote tallies are shown as recorded, and comments reflect the views of individual contributors.
From public community threads 2
-
taylz876 👍 1 Answered DWhy D is right • GPUs and pre-configured frameworks: Workbench instances can have GPUs attached and come with PyTorch preinstalled, so there's little to set up. • Secure BigQuery access without key management: the instance uses its managed identity (a service account attached to the instance), so nobody has to create or download a key, which is the more secure approach. • Least setup: it's a managed notebook environment. Why the others are wrong • A: A self-built JupyterHub needs driver installs, and having each person download service account keys is insecure and a lot of overhead. • B: A serverless training job isn't an interactive environment for prototyping. • C: Colab Enterprise is for live co-editing, but the "free version" isn't the right fit, and a shared notebook isn't the secure, key-free setup the question wants.
-
n2183712847 👍 1 Answered DCorrect answer: D. Agent Platform Workbench provides managed notebook instances with GPU support and preinstalled ML frameworks, while managed identity allows secure BigQuery access without downloading or managing service account keys; A requires significant infrastructure setup and insecure key management, B is designed for running training jobs rather than collaborative interactive prototyping, and C does not provide the specified managed GPU-enabled development environment in the same way.
Reader comments & corrections
No reader corrections yet — spotted an error or have a note? Share it below.
Related Analysis
Practice All PMLE Questions
Access 65 questions with complete answers and detailed explanations.
View Full PMLE Practice Test →