Artificial Intelligence Machine Learning Syllabus

Machine Learning syllabus — full chapter and topic list

Expand any chapter to see its topics and sub-topics. This is the whole examinable outline for Machine Learning in Artificial Intelligence, not a summary of it.

1. 1

   Learning from Examples

   4 topics

   • Forms of Learning
   • Supervised Learning
   • Decision Trees
   • Linear and Logistic Regression
2. 2

   Model Evaluation and Selection

   4 topics

   • Training, Validation, and Test Sets
   • Cross-Validation
   • Bias-Variance Tradeoff
   • Regularization
3. 3

   Classical Learning Algorithms

   4 topics

   • K-Nearest Neighbors
   • Support Vector Machines
   • Naive Bayes Classifier
   • Ensemble Methods
4. 4

   Neural Networks and Deep Learning

   4 topics

   • Perceptrons and Multilayer Networks
   • Backpropagation
   • Convolutional Neural Networks
   • Recurrent Networks and Transformers
5. 5

   Unsupervised Learning

   3 topics

   • Clustering
   • Dimensionality Reduction
   • Learning with Hidden Variables and EM
6. 6

   Reinforcement Learning

   5 topics

   • Passive Reinforcement Learning
   • Active Reinforcement Learning
   • Temporal-Difference Learning
   • Function Approximation and Policy Search
   • Deep Reinforcement Learning

Machine Learning flashcards for Artificial Intelligence

25 of 51 cards from the Machine Learning deck — real questions with worked answers.

1. What are the three main forms of machine learning based on feedback available to the learner?

   Supervised learning (learns from labeled input-output pairs), unsupervised learning (finds structure in unlabeled data), and reinforcement learning (learns from rewards/penalties via interaction with an environment).

2. Define supervised learning and state its two principal task types.

   Learning a function $f: X \to Y$ from labeled training examples $\{(x_i, y_i)\}$. The two task types are classification (discrete $y$) and regression (continuous $y$).

3. In a decision tree, what criterion does the ID3/C4.5 algorithm use to select the splitting attribute?

   Information gain, which chooses the attribute that maximally reduces entropy: $IG(S, A) = H(S) - \sum_{v} \frac{|S_v|}{|S|} H(S_v)$.

4. Give the formula for the entropy of a set $S$ with class probabilities $p_i$.

   $$H(S) = -\sum_{i} p_i \log_2 p_i$$

5. Give the formula for the Gini impurity used by CART decision trees.

   $$G = 1 - \sum_{i=1}^{K} p_i^{2}$$ where $p_i$ is the fraction of class $i$ in the node.

6. What is the hypothesis (model) of linear regression for a feature vector $\vec{x}$?

   $$\hat{y} = \vec{w}^{\top}\vec{x} + b = \sum_{j} w_j x_j + b$$

7. What cost function does linear regression minimize, and what is the closed-form (normal equation) solution?

   It minimizes the mean squared error $J(\vec{w}) = \frac{1}{n}\sum_i (y_i - \hat{y}_i)^2$. Closed form: $\hat{\vec{w}} = (X^{\top}X)^{-1}X^{\top}\vec{y}$.

8. What is the logistic (sigmoid) function used in logistic regression, and what is its output range?

   $$\sigma(z) = \frac{1}{1 + e^{-z}}$$ with output in $(0, 1)$, interpreted as $P(y=1 \mid \vec{x})$.

9. What loss function does logistic regression minimize?

   The binary cross-entropy (log loss): $$J = -\frac{1}{n}\sum_{i} \left[ y_i \log \hat{y}_i + (1 - y_i)\log(1 - \hat{y}_i) \right]$$

10. Why can't linear regression's squared-error loss simply be reused for classification with a sigmoid output?

    Combining the sigmoid with squared error yields a non-convex cost surface with many local minima; cross-entropy restores convexity and better-behaved gradients.

11. Distinguish the training, validation, and test sets by their purpose.

    Training set: fits model parameters. Validation set: tunes hyperparameters and selects models. Test set: gives an unbiased final estimate of generalization; used only once, at the end.

12. What is data leakage in the context of dataset splitting?

    When information from the validation/test set (or the future) improperly influences training, e.g. scaling using statistics computed over the whole dataset, producing overoptimistic performance estimates.

13. Describe k-fold cross-validation.

    The data is split into $k$ equal folds; the model is trained on $k-1$ folds and validated on the remaining fold, repeated $k$ times so each fold serves once as validation. The score is averaged over the $k$ runs.

14. What is leave-one-out cross-validation (LOOCV) and its main drawback?

    A special case of $k$-fold with $k = n$ (each single example is a fold). It is nearly unbiased but computationally expensive and has high variance in its estimate.

15. When should stratified cross-validation be used?

    For classification with imbalanced classes; stratification preserves the class proportions in each fold so estimates are not distorted by uneven splits.

16. State the bias-variance decomposition of expected prediction error.

    $$\mathbb{E}[(y - \hat{f}(x))^2] = \text{Bias}[\hat{f}]^2 + \text{Var}[\hat{f}] + \sigma^2$$ where $\sigma^2$ is irreducible noise.

17. Define bias and variance in the bias-variance tradeoff.

    Bias is error from wrong model assumptions (underfitting); variance is error from sensitivity to fluctuations in the training set (overfitting). Reducing one typically increases the other.

18. How do model complexity and the bias-variance tradeoff relate?

    As complexity increases, bias decreases but variance increases. Optimal generalization occurs at the complexity minimizing total error (the U-shaped test-error curve).

19. Write the objective of L2 (ridge) regularization for linear regression.

    $$J(\vec{w}) = \sum_i (y_i - \hat{y}_i)^2 + \lambda \sum_j w_j^{2}$$

20. How does L1 (Lasso) regularization differ from L2 (Ridge) in effect?

    L1 penalizes $\sum_j |w_j|$ and drives some weights exactly to zero (feature selection / sparsity); L2 penalizes $\sum_j w_j^2$ and shrinks weights smoothly toward zero without eliminating them.

21. What does the regularization hyperparameter $\lambda$ control?

    The strength of the penalty: larger $\lambda$ increases bias but reduces variance (more shrinkage/simpler model); $\lambda = 0$ recovers the unregularized model.

22. How does the k-Nearest Neighbors algorithm classify a new point?

    It finds the $k$ closest training points (by a distance metric) and assigns the majority class among them (for regression, averages their values). It is a lazy, non-parametric method with no explicit training phase.

23. Give the Euclidean and Manhattan distance formulas used in kNN.

    Euclidean: $d = \sqrt{\sum_j (x_j - x'_j)^2}$. Manhattan: $d = \sum_j |x_j - x'_j|$.

24. What is the effect of choosing a very small vs very large $k$ in kNN?

    Small $k$ gives low bias but high variance (noisy, jagged boundaries); large $k$ gives high bias but low variance (smoother boundaries, may underfit).

25. What does a Support Vector Machine maximize, and what are support vectors?

    It maximizes the margin, the distance between the decision hyperplane and the nearest points. The support vectors are those closest training points that lie on the margin and define the boundary.

See more Machine Learning flashcards →

Planning Machine Learning for Artificial Intelligence

Machine Learning (Artificial Intelligence) FAQ