support@Free-Braindumps.com
Register Login
Cisco
CompTIA
ISC
EC-Council
EXIN
Fortinet
ITIL
Juniper
Microsoft
VMware
Avaya
SAP
Google
NVIDIA
Salesforce
Amazon
Palo Alto Networks
ISC2
Splunk
ServiceNow
All Vendors
  1. Home
  2. Exams
  3. Google
  4. PROFESSIONAL MACHINE LEARNING ENGINEER

Google PROFESSIONAL MACHINE LEARNING ENGINEER Exam Actual Questions
Professional Machine Learning Engineer (Page 9 )

Updated On: 5-Jul-2026
Page 8 of 44
PDF with 339 Questions

You work for an online travel agency that also sells advertising placements on its website to other companies. You have been asked to predict the most relevant web banner that a user should see next. Security is important to your company. The model latency requirements are 300ms@p99, the inventory is thousands of web banners, and your exploratory analysis has shown that navigation context is a good predictor. You want to Implement the simplest solution. How should you configure the prediction pipeline?

  1. Embed the client on the website, and then deploy the model on AI Platform Prediction.
  2. Embed the client on the website, deploy the gateway on App Engine, and then deploy the model on AI Platform Prediction.
  3. Embed the client on the website, deploy the gateway on App Engine, deploy the database on Cloud Bigtable for writing and for reading the user's navigation context, and then deploy the model on AI Platform Prediction.
  4. Embed the client on the website, deploy the gateway on App Engine, deploy the database on Memorystore for writing and for reading the user's navigation context, and then deploy the model on Google Kubernetes Engine.

Answer(s): C

Explanation:

Option C is correct because it provides a minimal, latency-conscious pipeline: client embedding on the site, an App Engine gateway to handle requests securely and with low latency, a fast in-memory/write-optimized store (Memorystore would be options, but here Bigtable is chosen in C for scalable navigation context storage), and a model hosted on AI Platform Prediction for inference. It satisfies security, scalability for thousands of banners, and p99 latency. A) Lacks a gateway, risking exposed endpoints and throttling. B) Adds an unnecessary gateway without a fast read/write store for navigation context. D) Misuses Memorystore for the large, persistent navigation context and favors Kubernetes, increasing complexity and latency. Incorrect: A, B, D



Your team is building a convolutional neural network (CNN)-based architecture from scratch. The preliminary experiments running on your on-premises CPU-only infrastructure were encouraging, but have slow convergence. You have been asked to speed up model training to reduce time-to-market. You want to experiment with virtual machines (VMs) on Google Cloud to leverage more powerful hardware. Your code does not include any manual device placement and has not been wrapped in Estimator model-level abstraction.
Which environment should you train your model on?

  1. A VM on Compute Engine and 1 TPU with all dependencies installed manually.
  2. A VM on Compute Engine and 8 GPUs with all dependencies installed manually.
  3. A Deep Learning VM with an n1-standard-2 machine and 1 GPU with all libraries pre-installed.
  4. A Deep Learning VM with more powerful CPU e2-highcpu-16 machines with all libraries pre-installed.

Answer(s): C

Explanation:

Option C is correct because a Deep Learning VM provides pre-installed ML libraries and GPU support out of the box, enabling faster setup and leveraging single-GPU acceleration suitable for non-Estimator, non-device-placement code. It reduces setup overhead and aligns with on-prem CPU-to-GPU acceleration needs without requiring manual dependency management.
A) Incorrect — manual TPU setup is unnecessary and costly for CNN training without device placement; TPUs require special libraries and orchestration not ideal for this scenario.
B) Incorrect — 8 GPUs with manual setup increases complexity and cost; for a non-Estimator workflow, a single GPU on a DL VM is typically sufficient to improve convergence.
D) Incorrect — CPU-only Deep Learning VMs do not provide GPU acceleration, which is essential for faster training.



You work on a growing team of more than 50 data scientists who all use AI Platform. You are designing a strategy to organize your jobs, models, and versions in a clean and scalable way. Which strategy should you choose?

  1. Set up restrictive IAM permissions on the AI Platform notebooks so that only a single user or group can
    access a given instance.
  2. Separate each data scientist's work into a different project to ensure that the jobs, models, and versions created by each data scientist are accessible only to that user.
  3. Use labels to organize resources into descriptive categories. Apply a label to each created resource so that users can filter the results by label when viewing or monitoring the resources.
  4. Set up a BigQuery sink for Cloud Logging logs that is appropriately filtered to capture information about AI Platform resource usage. In BigQuery, create a SQL view that maps users to the resources they are using

Answer(s): C

Explanation:

Option C is correct because labels provide a scalable, non-intrusive way to organize and filter a large collection of AI Platform resources (jobs, models, versions) across many users without altering IAM, project structure, or log sinks. They support per-resource metadata and powerful querying/monitoring in a single view across the enterprise.
A) Incorrect — Restrictive IAM on notebooks is not scalable for organizing resources; it governs access to notebooks, not the broader resource hierarchy or labeling and filtering.
B) Incorrect — Creating a separate project per data scientist is not scalable for 50+ users and fragments resources; it complicates management and cross-user collaboration.
D) Incorrect — Logging sinks and SQL views add complexity and latency; labels are the recommended, native mechanism for organization and filterability.



You are training a deep learning model for semantic image segmentation with reduced training time. While using a Deep Learning VM Image, you receive the following error: The resource 'projects/ deeplearning-platform/zones/europe-west4-c/acceleratorTypes/nvidia-tesla-k80' was not found. What should you do?

  1. Ensure that you have GPU quota in the selected region.
  2. Ensure that the required GPU is available in the selected region.
  3. Ensure that you have preemptible GPU quota in the selected region.
  4. Ensure that the selected GPU has enough GPU memory for the workload.

Answer(s): B

Explanation:

Option B is correct because the error indicates the specific accelerator type (nvidia-tesla-k80) is not available in the chosen region, so ensuring the required GPU is available in that region aligns with resolving a regional availability constraint.
A) Incorrect — GPU quota is about the amount of resources you can allocate, not the actual availability of that specific accelerator type in the region, which is the reported issue.
C) Incorrect — Preemptible GPU quota concerns cost and interruption timing, not the fundamental availability of the requested GPU type in the region.
D) Incorrect — Memory capacity is a workload suitability concern, but the error refers to nonexistence/availability of the accelerator type, not insufficient memory.



Your team is working on an NLP research project to predict political affiliation of authors based on articles they have written. You have a large training dataset that is structured like this:



You followed the standard 80%-10%-10% data distribution across the training, testing, and evaluation subsets.

How should you distribute the training examples across the train-test-eval subsets while maintaining the 80-10- 10 proportion?

  1. Distribute texts randomly across the train-test-eval subsets:
    Train set: [TextA1, TextB2, ...]
    Test set: [TextA2, TextC1, TextD2, ...]
    Eval set: [TextB1, TextC2, TextD1, ...]
  2. Distribute authors randomly across the train-test-eval subsets: (*) Train set: [TextA1, TextA2, TextD1, TextD2, ...]
    Test set: [TextB1, TextB2, ...]
    Eval set: [TexC1,TextC2 ...]
  3. Distribute sentences randomly across the train-test-eval subsets:
    Train set: [SentenceA11, SentenceA21, SentenceB11, SentenceB21, SentenceC11, SentenceD21 ...] Test set: [SentenceA12, SentenceA22, SentenceB12, SentenceC22, SentenceC12, SentenceD22 ...] Eval set: [SentenceA13, SentenceA23, SentenceB13, SentenceC23, SentenceC13, SentenceD31 ...]
  4. Distribute paragraphs of texts (i.e., chunks of consecutive sentences) across the train-test-eval subsets:
    Train set: [SentenceA11, SentenceA12, SentenceD11, SentenceD12 ...] Test set: [SentenceA13, SentenceB13, SentenceB21, SentenceD23, SentenceC12, SentenceD13 ...] Eval set: [SentenceA11, SentenceA22, SentenceB13, SentenceD22, SentenceC23, SentenceD11 ...]

Answer(s): B

Explanation:

Option B is correct because randomizing at the author level preserves author-specific style and prevents leakage of author identity across splits, which is essential for generalization in author-prediction tasks. A) Randomly across all texts risks splitting a single author’s works between train/test/eval, causing data leakage. C) Random sentence-level splits break document-context and can still leak author signals across subsets. D) Splitting by paragraphs/chunks can still leak author/style signals if authors appear in multiple chunks across subsets. Therefore, author-level stratification maintains 80-10-10 proportions while preventing cross-set contamination.
A) Incorrect — Random text-level split risks leakage.
B) Correct — See above.
C) Incorrect — Sentence-level splits can leak author signals.
D) Incorrect — Paragraph-level chunks may still leak author identity.



Your team has been tasked with creating an ML solution in Google Cloud to classify support requests for one of your platforms. You analyzed the requirements and decided to use TensorFlow to build the classifier so that you have full control of the model's code, serving, and deployment. To save time, you want to build on existing resources and use managed services instead of building a completely new model. How should you build the classifier?

  1. Use the Natural Language API to classify support requests.
  2. Use AutoML Natural Language to build the support requests classifier.
  3. Use an established text classification model on Vertex AI to perform transfer learning.
  4. Use an established text classification model on Vertex AI as-is to classify support requests.

Answer(s): C

Explanation:

Option C is correct because transfer learning from an established text classification model on Vertex AI leverages pre-trained linguistic representations while allowing fine-tuning on domain-specific support requests, giving you control over deployment and code with managed infrastructure.
A) Incorrect — Natural Language API provides predefined classifiers but does not offer transfer learning or full model control for custom deployment.
B) Incorrect — AutoML Natural Language automates model training but abstracts away model code and full control required for your TensorFlow-based solution.
D) Incorrect — Using the model as-is lacks domain adaptation and the ability to customize for your specific classes and deployment needs.



You recently joined a machine learning team that will soon release a new project. As a lead on the project, you are asked to determine the production readiness of the ML components. The team has already tested features and data, model development, and infrastructure. Which additional readiness check should you recommend to the team?

  1. Ensure that training is reproducible.
  2. Ensure that all hyperparameters are tuned.
  3. Ensure that model performance is monitored.
  4. Ensure that feature expectations are captured in the schema.

Answer(s): C

Explanation:

Option C is correct because continuous model performance monitoring is essential for production readiness to detect drift, degradation, and anomalies in real time. A, While reproducibility is important for development, it’s already part of model training pipelines and not specifically a production readiness check. B, Hyperparameter tuning is a development optimization task, not a production readiness metric. D, Capturing feature expectations in the schema is valuable for data contracts but does not address ongoing production performance monitoring requirements.



You work for a credit card company and have been asked to create a custom fraud detection model based on historical data using AutoML Tables. You need to prioritize detection of fraudulent transactions while minimizing false positives. Which optimization objective should you use when training the model?

  1. An optimization objective that minimizes Log loss
  2. An optimization objective that maximizes the Precision at a Recall value of 0.50
  3. An optimization objective that maximizes the area under the precision-recall curve (AUC PR) value
  4. An optimization objective that maximizes the area under the receiver operating characteristic curve (AUC ROC) value

Answer(s): C

Explanation:

Option C is correct because maximizing area under the precision-recall curve (AUC PR) aligns with high fraud detection under class-imbalanced data, emphasizing precision at relevant recall ranges and reducing false positives while maintaining detection rate.
A) Incorrect — minimizing Log loss optimizes overall probabilistic calibration, not specifically the balance of detection vs. false positives at fraud-heavy costs.
B) Incorrect — maximizing Precision at a fixed Recall (0.50) is a threshold-specific metric and not the AUC objective; it may not capture performance across varying fraud rates.
D) Incorrect — AUC ROC can be misleading with imbalanced data, as it weights negatives heavily and may underemphasize the positive class performance critical for fraud.



  PDF
Viewing page 9 of 44
Viewing questions 65 - 72 out of 339 questions


Post your Comments and Discuss Google PROFESSIONAL MACHINE LEARNING ENGINEER exam prep with other Community members:

AI Tutor AI Tutor 👋 I’m here to help!