1. What is the difference between AI and ML?
- AI concentrates on algorithms while ML is about theory
- AI is ML but without mathematics
- AI is a discipline while ML is a toolset
- AI and ML are the same
Correct answer: ✅ AI is a discipline while ML is a toolset
Explanation (clear & exam-friendly)
Artificial Intelligence (AI) is the broader discipline focused on creating systems that can mimic human intelligence (reasoning, planning, perception, decision-making).
Machine Learning (ML) is a subset/toolset within AI that enables systems to learn patterns from data and improve performance without being explicitly programmed.
In short:
AI = the goal and field of study
ML = one of the primary methods used to achieve AI
Why the other options are incorrect ❌
❌ AI concentrates on algorithms while ML is about theory
Incorrect. ML is highly algorithmic and mathematical, and AI includes theory, systems, and applications.
❌ AI is ML but without mathematics
Incorrect. AI heavily relies on mathematics (logic, probability, optimization), and ML is deeply mathematical.
❌ AI and ML are the same
Incorrect. ML is a subset of AI, not synonymous with it.
Quick memory rule 🧠
All ML is AI, but not all AI is ML.
2. What is the primary impact of ML?
- It allows business operations to scale
- It allows businesses to be more accurate in their predictions
- Provides insights that were not previously possible
- Cost savings
Correct answer: ✅ Provides insights that were not previously possible
Explanation (exam-oriented and conceptually correct)
The primary impact of Machine Learning (ML) is its ability to discover complex patterns and relationships in data that were previously too difficult or impossible to identify with traditional rule-based or statistical systems.
ML enables:
Detection of non-obvious patterns
Learning from large, high-dimensional datasets
Generating new insights rather than just automating existing logic
Everything else listed is typically a secondary or downstream benefit.
Why the other options are not the primary impact ❌
❌ It allows business operations to scale
Scaling is mainly driven by cloud infrastructure and automation, not ML itself.
❌ It allows businesses to be more accurate in their predictions
True, but this is a result of the deeper insights ML uncovers, not the fundamental impact.
❌ Cost savings
Cost reduction is an indirect outcome, often achieved after insights improve efficiency or decisions.
3. True or False? Most business data is unstructured data, and mainly text.
- True
- False
Correct answer:✅ True
Correct Explanation (industry & exam aligned)
Most business data today is unstructured, and a large portion of it is text-based.
Examples include:
Emails
Chat messages
Customer reviews
Support tickets
Documents (PDFs, Word files)
Social media posts
Logs and free-form text fields
Industry estimates (commonly cited by IBM, Gartner, and others) consistently state that ~70–90% of enterprise data is unstructured.
While structured data (tables, transactions, metrics) is critical and heavily used for analytics, it represents a smaller portion by volume compared to unstructured data.
Why the earlier assumption can be misleading
Data engineers work more often with structured data, so it feels dominant.
ML and AI discussions, however, correctly emphasize that most raw business data is unstructured.
Exams sometimes test data reality, not just engineering workflows.
Final Answer (corrected)
True ✅
Most business data is unstructured data, and mainly text.
Exam memory rule 🧠
By volume → Mostly unstructured
By analytics usage → Mostly structured
4. Google Cloud's pretrained model APIs use:
- Your models and your data
- Google's models and your data
- Google's models and Google's data
- Your models and Google's data
Correct answer: ✅ Google’s models and your data
Explanation (exam-accurate)
Google Cloud pretrained model APIs (such as Vision API, Natural Language API, Speech-to-Text, Translation, etc.) work as follows:
Models → Built, trained, and maintained by Google
Data → You provide your own input data (images, text, audio, video) at inference time
You do not train or manage the models—you simply send requests and receive predictions.
Why the other options are incorrect ❌
❌ Your models and your data
This describes custom ML models (e.g., Vertex AI custom training), not pretrained APIs.
❌ Google’s models and Google’s data
Google does not use its own proprietary data to make predictions for your requests.
Your input data is used only to generate results for you.
❌ Your models and Google’s data
This scenario does not apply to any Google Cloud ML offering.
5. Select the statement that does not apply to Notebooks.
They use the latest open-source version of JupyterLab
You can easily change hardware including adding and removing GPUs
Notebook instances are standard Compute Engine instances that live in your projects
It's up to you to install the latest ML libraries on the notebooks.
Correct answer: ❌ It’s up to you to install the latest ML libraries on the notebooks
Explanation (GCP ML / PDE exam–accurate)
Vertex AI Workbench (formerly AI Platform Notebooks) comes with prebuilt, managed notebook images that already include:
Popular ML frameworks (TensorFlow, PyTorch, scikit-learn)
Common data science libraries
GPU / CUDA support (when applicable)
Google maintains and updates these images, so it is not solely your responsibility to install or manage the latest ML libraries.
Why the other statements DO apply ✅
✅ They use the latest open-source version of JupyterLab
Vertex AI Workbench is based on JupyterLab, regularly updated by Google.
✅ You can easily change hardware including adding and removing GPUs
Notebook instances run on Compute Engine, and you can stop/start to:
Change machine type
Add or remove GPUs
✅ Notebook instances are standard Compute Engine instances that live in your projects
They are GCE VMs inside your own GCP project, with full IAM and VPC control.
6. True or False? Notebooks contain a magic function to execute BigQuery
- False
- True
Correct answer: ✅ True
Explanation (exam-accurate)
GCP Notebooks (for example, Vertex AI Workbench) include built-in Jupyter “magic” commands that allow you to run BigQuery SQL directly from a notebook cell.
Examples include:
%%bigquery (cell magic)
%bigquery (line magic)
These magics let you:
Execute SQL against BigQuery
Automatically load results into pandas DataFrames
Analyze data interactively without writing client boilerplate code
PDE / ML exam memory rule 🧠
GCP Notebooks + BigQuery = built-in BigQuery magic available
So the statement
“Notebooks contain a magic function to execute BigQuery”
is True ✅.
7. TensorFlow Hub has templates for which of the following?
- All other answers are correct
- Trained models
- Kubeflow pipelines and components
- Jupyter notebooks
Correct answer: ✅ All other answers are correct
Explanation (exam-accurate)
TensorFlow Hub provides reusable assets and templates to help teams quickly build ML solutions. These include:
✅ Trained models
Pretrained and fine-tunable models for NLP, vision, audio, etc.
✅ Jupyter notebooks
Example notebooks showing how to load, fine-tune, and use models from TensorFlow Hub.
✅ Kubeflow pipelines and components
Reference pipelines/components (often linked via associated repos) to operationalize models in production ML workflows.
8. Which technology was developed as a solution to run Kubernetes clusters and pods behind the scenes to support deploying pipelines?
- Cloud Orchestrator
- Vertex Pipelines
- Kubeflow
- Cloud Composer
Correct answer: ✅ Vertex Pipelines
Why Vertex Pipelines is the correct answer
The question asks:
Which technology was developed as a solution to run Kubernetes clusters and pods behind the scenes to support deploying pipelines?
The key phrase here is “behind the scenes.”
Vertex AI Pipelines was specifically created to:
Abstract away Kubernetes
Run Kubeflow Pipelines without requiring users to manage GKE clusters
Automatically handle:
Kubernetes clusters
Pods
Scaling
Infrastructure lifecycle
From a user perspective, you do not see or manage Kubernetes at all — it is fully managed by Google Cloud.
That is exactly what the question is testing.
Why the other options are incorrect ❌
❌ Kubeflow
Kubeflow requires you to run and manage Kubernetes (GKE)
You must understand clusters, nodes, pods
Kubernetes is not hidden
Kubeflow is the foundation, not the “behind-the-scenes” solution
❌ Cloud Composer
Managed Apache Airflow
Used for data orchestration, not Kubernetes-based ML pipelines
❌ Cloud Orchestrator
Not a valid Google Cloud product
9. BigQuery ML has support for which of the following modeling tasks:
- Clustering
- Computer vision
- Classification
- Regression
Correct selections (✔):
✅ Clustering
✅ Classification
✅ Regression
Explanation (exam-accurate)
BigQuery ML enables you to train and run ML models directly using SQL inside BigQuery. It supports several tabular-data ML tasks, including:
Classification
Binary and multi-class classification
Examples: churn prediction, fraud detection
Regression
Linear and boosted tree regression
Examples: sales forecasting, price prediction
Clustering
k-means clustering
Examples: customer segmentation
These are all core BigQuery ML capabilities and are frequently tested.
Why the remaining option is incorrect ❌
❌ Computer vision
BigQuery ML does not train or run computer vision models
Vision tasks (image/video) are handled by:
Vertex AI
Vision API
AutoML Vision
BigQuery ML is optimized for structured/tabular data, not images or video.
10. True or False? You can train and evaluate machine learning models directly in BigQuery.
- False
- True
Correct answer: ✅ True
Explanation (exam-accurate)
With BigQuery ML, you can train, evaluate, and make predictions directly in BigQuery using SQL—without exporting data to external ML frameworks.
BigQuery ML supports:
CREATE MODEL → train models
ML.EVALUATE → evaluate models
ML.PREDICT → run predictions
All of this runs inside BigQuery, making it ideal for analysts and data engineers.
PDE / ML exam memory rule 🧠
BigQuery ML = ML with SQL, no data movement required
So the statement
“You can train and evaluate machine learning models directly in BigQuery”
is True ✅.
11. Which of the following are valid techniques for improving AutoML Vision and AutoML Natural Language models?
Increase the diversity and complexity of data
Increase the number of labels
Increase the amount of training data
Ensure consistent labeling
Correct selections (✔):
✅ Increase the diversity and complexity of data
✅ Increase the amount of training data
✅ Ensure consistent labeling
Explanation (exam-accurate)
For AutoML Vision and AutoML Natural Language (part of Vertex AI), model quality is driven primarily by data quality and representativeness.
✅ Increase the diversity and complexity of data
Helps the model generalize better
Covers real-world variations (lighting, language style, formats, edge cases)
Reduces overfitting
👉 Strongly recommended
✅ Increase the amount of training data
More examples → better pattern learning
Especially important for rare classes
👉 One of the most impactful improvements
✅ Ensure consistent labeling
Inconsistent or noisy labels directly degrade model accuracy
AutoML assumes labels are correct and consistent
Label quality often matters more than quantity
👉 Critical best practice
Why the remaining option is NOT necessarily correct ❌
❌ Increase the number of labels
Simply adding more labels does not improve model quality
Can actually:
Increase confusion
Require much more data per label
Only useful if the new labels are meaningful and well-represented
12. AutoML makes use of which of the following:
Your models and Google's data
Google's models and Google's data
Google's models and your data
Your models and your data
Correct answer: ✅ Google’s models and your data
Explanation (exam-accurate)
AutoML (part of Vertex AI) works by:
Using Google’s pre-designed and optimized model architectures
Training them on your labeled data
You do not design the model architecture yourself, and Google does not use its own proprietary data to train models for your use case.
Why the other options are incorrect ❌
❌ Your models and Google’s data
This guarantees doesn’t happen.
Google never trains your models using its private data.
❌ Google’s models and Google’s data
That describes Google’s internal systems, not AutoML for customers.
❌ Your models and your data
This describes custom model training (e.g., custom training on Vertex AI), not AutoML.
