AIF Practitioner Exam - 1 Line Revision Sheet
Machine Learning Basics
- Ensemble Learning → Combine multiple models to improve accuracy.
- Gradient Descent → Algorithm that minimizes the error (loss function).
- Epoch → One complete pass through the training data.
- Curse of Dimensionality → Too many features make learning harder.
- Normalization → Scale features so all contribute equally.
- Imputation → Fill missing values in data.
- Validation Set → Used for hyperparameter tuning.
- Confusion Matrix → Shows actual vs predicted classifications.
- ROC Curve → Trade-off between sensitivity and specificity.
- Activation Function → Adds non-linearity to neural networks.
Machine Learning Types
- Supervised Learning → Learn from labeled data.
- Unsupervised Learning → Find patterns without labels.
- Reinforcement Learning → Learn by rewards and penalties.
- Classification → Predict categories.
- Regression → Predict numbers.
- Clustering → Group similar records.
Algorithms
- Decision Tree → Used for classification and regression.
- Random Forest → Multiple decision trees combined.
- Gradient Boosting/XGBoost → Powerful ensemble for structured data.
- K-Means → Clustering algorithm.
- Naive Bayes → Probability-based classifier.
- Neural Network → Learns complex patterns using layers.
Model Quality
- Good Model → Low Bias + Low Variance.
- Imbalanced Dataset → Use Oversampling, Undersampling, or SMOTE.
- Precision-Recall Curve → Best metric for imbalanced data.
Amazon Bedrock
- Amazon Bedrock → Managed service for foundation models.
- Bedrock Runtime API → Used for inference requests.
- Bedrock Agent Runtime API → Invoke agents and knowledge bases.
- Agents → Execute multi-step tasks.
- Playgrounds → Test prompts and model settings.
- Guardrails → Block harmful or unwanted responses.
- Bedrock Studio → Rapid prototyping environment.
- Model Customization → Creates private customized model copies.
- Fine-Tuning → Train model with domain-specific data.
- Grounding/RAG → Reduces hallucinations using enterprise data.
Amazon Q
- Amazon Q Business → Enterprise AI assistant.
- Q Business Responses → Based on company data and permissions.
- Q Apps → Create and share AI-powered apps.
- Guardrails + Keywords → Control Q Business behavior.
- RAG in Q Business → Provides factual and traceable answers.
SageMaker
- Data Wrangler → Data preparation and feature engineering.
- Feature Store → Central repository for ML features.
- Ground Truth → Data labeling service.
- JumpStart → Pre-trained models and solutions.
- Studio Lab → Free ML experimentation environment.
- Debugger → Real-time training diagnostics.
- Experiments → Track and compare ML experiments.
- Clarify → Detect bias and explain model predictions.
AWS AI Services
- Rekognition → Image and video analysis.
- Textract → Extract text from documents.
- Translate → Language translation.
- Polly → Text-to-speech.
- Lex → Build chatbots.
Responsible AI
- Fairness → Treat users equally.
- Privacy → Protect personal data.
- Safety → Prevent harmful outputs.
- Explainability → Understand model decisions.
- Transparency → Be clear about AI usage.
- AI Service Cards → Explain AI capabilities and limitations.
- Watermarking → Identify AI-generated content.
Prompt Engineering
- Context → Relevant information improves answers.
- Specific Instructions → Better prompts = better results.
- Feedback Loop → Continuously improve prompts.
- Prompt Robustness → Use dynamic, context-aware templates.
AWS Security
- IAM Identity Center → Secure enterprise access.
- AWS Shield → DDoS protection.
- Step Functions → Orchestrate AI/ML workflows.
- Lambda → Run serverless AI tasks.
Last Minute Exam Crib Sheet (20 Must Remember)
- Ensemble = Multiple models.
- Gradient Descent = Minimize loss.
- Epoch = One full training pass.
- Classification = Predict category.
- Regression = Predict number.
- K-Means = Clustering.
- Decision Tree = Classification/Regression.
- Validation Set = Tune hyperparameters.
- ROC = Sensitivity vs Specificity.
- Data Wrangler = Prepare data.
- Feature Store = Store features.
- Ground Truth = Label data.
- JumpStart = Prebuilt models.
- Debugger = Training diagnostics.
- Clarify = Bias detection.
- Bedrock Runtime = Inference API.
- Agents = Multi-step tasks.
- Guardrails = Safe responses.
- RAG = Reduce hallucinations.
- Q Business = Enterprise AI assistant.
