AML Cheat Sheet 2026

The 30 highest-yield AML facts, distilled from real exam questions. Print it, save it as a PDF, or study it here — free, no sign-up.

1. Which technique addresses the problem of imbalanced classes specifically during the training of a decision tree? → Using 'class_weight' parameter to penalize majority class errors more
2. What effect does setting the discount factor (γ) close to 0 have on a reinforcement learning agent? → The agent becomes myopic, prioritizing only immediate rewards
3. What does the Bellman equation fundamentally express in reinforcement learning? → The recursive relationship between a state's value and the values of its successor states
4. What does a log transformation primarily help with during feature engineering? → Reducing the skewness of right-skewed distributions
5. What is the primary purpose of stress-testing an ML model during risk assessment? → To evaluate model behavior under extreme or adversarial input conditions
6. When a model's decision is challenged in a high-stakes context, the practitioner's ethical duty of 'explainability' MOST directly serves which goal? → Enabling affected individuals to understand and contest the decision
7. In a confusion matrix for a binary classifier, which metric is computed as TP / (TP + FP)? → Precision
8. Which neural network is best suited for sequential data? → Recurrent Neural Network
9. In a decision tree, information gain is computed as the reduction in which quantity after a split? → Entropy
10. When assessing risks of an ML model in a regulated industry, which document type formally captures identified risks and corresponding mitigations? → A model risk register
11. Which method identifies predictive features by measuring how much each reduces impurity in a trained Random Forest? → Feature importance from Random Forest
12. What does accountability in AI mean? → Ensuring responsibility for AI impacts
13. In AML practice, what is the purpose of a standard operating procedure (SOP)? → To document step-by-step instructions for routine tasks to ensure consistency and quality
14. Which NLP technique converts words into dense vector representations that capture semantic relationships? → Word2Vec / Word Embeddings
15. Which property of sinusoidal positional encodings in the original transformer makes them potentially generalizable to sequence lengths unseen during training? → They use fixed frequencies that can represent any integer position deterministically
16. What is the purpose of a Q-Q (quantile-quantile) plot in data analysis? → Assessing whether data follows a theoretical distribution
17. What is the purpose of cross-validation in model evaluation reporting? → To provide an unbiased estimate of model performance on unseen data
18. Which method involves searching through a range of hyperparameters? → Grid search
19. A machine learning team is designing a system to predict recidivism risk for parole decisions. The MOST ethically critical design requirement is: → Ensuring the system provides decision support, not autonomous binding decisions
20. Which of the following is a core requirement of the Colorado AI Act (SB 21-169) for high-risk AI systems? → Conducting impact assessments and providing adverse action notices
21. What does 'algorithmic bias' refer to? → Unfair outcomes produced by models
22. In the LSTM architecture, which gate controls how much of the previous cell state is retained? → Forget gate
23. Which technique addresses class imbalance during model evaluation by computing metrics on a resampled dataset that reflects equal class distribution? → Balanced accuracy metric on original data
24. When comparing two ML models using A/B testing in production, the minimum required sample size is determined by: → Desired statistical power, significance level, and minimum detectable effect size
25. What is the purpose of max pooling in a convolutional neural network? → Reduce spatial dimensions while retaining dominant features
26. Which fairness metric requires that the proportion of positive predictions be equal across protected groups, regardless of actual base rates? → Demographic parity
27. Which evaluation protocol is most appropriate for measuring a recommender system's performance when the goal is relevance of the top-N items shown to users? → Normalized Discounted Cumulative Gain (NDCG@N)
28. In ML project execution, what does 'data versioning' primarily help prevent? → Silent dataset drift causing irreproducible experiments when the underlying data changes
29. A client requests that an ML practitioner build a system to infer employees' political beliefs from social media data. The practitioner should FIRST: → Refuse and explain the serious ethical and legal risks of such profiling
30. What is the BLEU score used to measure in NLP? → Quality of machine-generated text by comparing n-gram overlap with reference translations