7.2. Common Exam Patterns and Distractors

What breaks without knowing these traps? You'll fall for the same tricks that catch everyone else. Imagine seeing "logistic regression" and thinking it predicts numbers because it has "regression" in the name—wrong, it's classification. Or seeing "extract text from scanned documents" and choosing NLP—wrong, the input is an image, so it's Computer Vision.

Consider these patterns like a cheat sheet for distractor avoidance. For instance, whenever a question mentions "bounding boxes" or "coordinates," the answer involves object detection—not classification. When you see "grouping by similar attributes without predefined categories," that's clustering (unsupervised), not classification (supervised). Learn these patterns, and you'll spot the traps before they catch you.

Pattern 1: Input Modality Confusion

The exam tests whether you classify AI workloads by INPUT type, not output type.

• Know: OCR is COMPUTER VISION (input is image), not NLP
• Know: Handwritten text extraction is COMPUTER VISION, not NLP
• Apply the framework: "What goes IN?" determines the workload category

Pattern 2: Logistic Regression Trap

Despite its name, logistic regression is CLASSIFICATION.

• Know: Logistic regression predicts categories (yes/no), not numbers
• Know: If predicting probability of an event (yes/no outcome), that's classification
• Remember: Regression predicts continuous numeric values

Pattern 3: Responsible AI Principle Matching

Questions present scenarios and ask which principle applies.

• Fairness = discrimination/bias based on gender, ethnicity, race
• Transparency = users understanding AI limitations
• Accountability = meeting legal/ethical standards
• Privacy & Security = protecting sensitive data (especially healthcare)
• Know: Salary, credit history, payment history are LEGITIMATE business factors, not fairness issues

Pattern 4: Azure Service Capabilities

Know what each service can and cannot do:

• Azure Translator = text-to-text ONLY (no speech)
• DALL-E = generates images, CANNOT describe/analyze images
• Azure Vision specialized domains = celebrities and landmarks ONLY

Pattern 5: Supervised vs Unsupervised

Classification by label presence:

• Labels present → Supervised (regression, classification)
• No labels → Unsupervised (clustering)
• "Grouping by similar attributes" without predefined categories → Clustering

Pattern 6: Features vs Labels

• Features = inputs (descriptive attributes)
• Labels = outputs (what you're predicting)
• If asked "how many features and labels," count: inputs → features, output → label