AI, ML, and Deep Learning Concepts

1. AI vs ML vs Deep Learning

These terms are often confused but represent nested concepts with increasing specificity.

Relationship:

  Artificial Intelligence (broadest)
    └─ Machine Learning
         └─ Deep Learning
              └─ Generative AI (e.g., LLMs)

2. Types of Machine Learning

3. ML Workflow on AWS

ML Workflow:

1. COLLECT DATA
   S3, Kinesis, DMS, Glue ETL, Data Lake

2. PREPARE DATA
   Glue DataBrew (clean), SageMaker Data Wrangler (transform)
   Label: SageMaker Ground Truth

3. BUILD MODEL
   SageMaker: built-in algorithms, custom code, notebooks
   Choose: algorithm, hyperparameters, training data

4. TRAIN MODEL
   SageMaker Training Jobs: managed GPU/CPU instances
   Distributed training for large models

5. EVALUATE MODEL
   Metrics: accuracy, precision, recall, F1, AUC
   SageMaker Model Monitor: detect drift

6. DEPLOY MODEL
   SageMaker Endpoints: real-time inference
   SageMaker Batch Transform: batch inference
   SageMaker Serverless Inference: pay per request

7. MONITOR
   SageMaker Model Monitor: data drift, model quality

4. Key ML Concepts for the Exam

Exam Tip

ML Concepts: "Learn from labeled data" = Supervised. "Find patterns in unlabeled data" = Unsupervised. "Trial-and-error with rewards" = Reinforcement. "Model too specific to training data" = Overfitting. "Model too simple" = Underfitting. "Label training data" = SageMaker Ground Truth. "Tune settings before training" = Hyperparameter tuning.