Innovating with Google Cloud AI

Artificial Intelligence (AI) is the broad field of creating machines that can perform tasks that typically require human intelligence. Machine Learning (ML) is a subset of AI where systems learn from data rather than being explicitly programmed. Deep Learning (DL) is a subset of ML that uses neural networks with many layers.

  +-------------------------------------------+
  |           Artificial Intelligence          |
  |   +-----------------------------------+   |
  |   |        Machine Learning           |   |
  |   |   +---------------------------+   |   |
  |   |   |      Deep Learning        |   |   |
  |   |   |  (Neural Networks, LLMs)  |   |   |
  |   |   +---------------------------+   |   |
  |   +-----------------------------------+   |
  +-------------------------------------------+
        

Types of Machine Learning

Key ML Concepts for the Exam

Google Cloud offers ML solutions at every level of expertise and customization. The exam frequently asks you to choose the right level for a given scenario. The spectrum ranges from no ML expertise needed (pre-trained APIs) to full custom training (Vertex AI with TensorFlow/PyTorch).

Google Cloud provides pre-trained ML models accessible via simple REST APIs. No ML expertise, training data, or model management is needed. Just send your data and receive predictions.

# Example: Call Cloud Vision API with gcloud
gcloud ml vision detect-labels gs://my-bucket/photo.jpg

# Example: Analyze sentiment with NL API
gcloud ml language analyze-sentiment \
  --content="Google Cloud is fantastic for data analytics!"

# Example: Transcribe audio
gcloud ml speech recognize gs://my-bucket/audio.wav \
  --language-code=en-US

AutoML lets you train custom ML models by simply providing your labeled data. Vertex AI AutoML handles feature engineering, architecture search, hyperparameter tuning, and model selection automatically. No ML coding expertise required.

AutoML Workflow

1. Prepare data — Upload labeled data (images with categories, text with labels, CSVs with target columns).
2. Train model — Vertex AI automatically tries thousands of architectures and selects the best one.
3. Evaluate — Review precision, recall, F1 score, and confusion matrix in the console.
4. Deploy — Deploy the model to an endpoint for online predictions, or run batch predictions.

AutoML Data Types

BigQuery ML (BQML) lets you create and execute ML models directly in BigQuery using standard SQL. This democratizes ML by enabling SQL analysts to build models without moving data out of the warehouse or learning Python.

-- Create a logistic regression model to predict customer churn
CREATE OR REPLACE MODEL `my_dataset.churn_model`
OPTIONS(
  model_type='LOGISTIC_REG',
  input_label_cols=['churned'],
  auto_class_weights=TRUE
) AS
SELECT
  tenure_months,
  monthly_charges,
  contract_type,
  internet_service,
  churned
FROM `my_dataset.customer_data`
WHERE split_col = 'TRAIN';

-- Evaluate the model
SELECT * FROM ML.EVALUATE(MODEL `my_dataset.churn_model`);

-- Make predictions on new data
SELECT * FROM ML.PREDICT(MODEL `my_dataset.churn_model`,
  (SELECT * FROM `my_dataset.new_customers`));

Supported Model Types in BQML

Vertex AI is Google Cloud's unified ML platform that brings together all ML services under one roof. It supports the entire ML lifecycle: data preparation, training (AutoML or custom), evaluation, deployment, monitoring, and MLOps.

Key Vertex AI Components

Generative AI on Vertex AI

Vertex AI provides access to Google's foundation models through the Generative AI Studio and Model Garden. Key models include:

• Gemini — Google's most capable multimodal model. Understands text, images, video, audio, and code.
• Imagen — Image generation and editing from text prompts.
• Codey — Code generation, completion, and explanation.
• Chirp — Advanced speech-to-text with high accuracy across languages.

TensorFlow is Google's open-source ML framework for building and training neural networks. TPUs (Tensor Processing Units) are Google's custom-designed chips optimized for ML workloads, providing significant speedups over GPUs for certain model architectures.

TPU vs GPU vs CPU

# Create a VM with a TPU for training
gcloud compute tpus tpu-vm create my-tpu \
  --zone=us-central1-b \
  --accelerator-type=v4-8 \
  --version=tpu-ubuntu2204-base

# Train a TensorFlow model on GPU via Vertex AI
gcloud ai custom-jobs create \
  --display-name=my-training-job \
  --worker-pool-spec=machine-type=n1-standard-8,\
    accelerator-type=NVIDIA_TESLA_T4,\
    accelerator-count=1,\
    replica-count=1,\
    container-image-uri=gcr.io/my-project/trainer:latest

Google has published AI Principles that guide responsible development and use of AI. The CDL exam tests awareness of these principles and their practical implications.

Google's AI Principles

1. Be socially beneficial — AI should benefit society broadly.
2. Avoid creating or reinforcing unfair bias — Test for and mitigate bias in data and models.
3. Be built and tested for safety — Rigorous testing and monitoring.
4. Be accountable to people — Human oversight and control.
5. Incorporate privacy design principles — Data minimization and user control.
6. Uphold high standards of scientific excellence — Peer review and reproducibility.
7. Be made available for uses that accord with these principles — Restrict harmful applications.

Explainable AI (XAI)

Vertex AI includes Explainable AI tools that help you understand why a model made a specific prediction. Feature attributions show which input features contributed most to the output. This is crucial for building trust, debugging models, and meeting regulatory requirements.

Quick Decision Matrix

• "Detect objects in images with standard labels" → Vision API
• "Classify customer emails into custom categories" → AutoML Text
• "Predict sales from BigQuery data using SQL" → BigQuery ML
• "Train a custom transformer on proprietary data" → Vertex AI Custom Training
• "Build a chatbot" → Dialogflow (simple) or Vertex AI + Gemini (advanced)
• "Generate text/images/code" → Vertex AI Generative AI / Gemini
• "What platform unifies all ML?" → Vertex AI
• "What accelerator is unique to Google?" → TPU