CISC 271 · Linear Data Analysis · Winter 2026
Queen's University · School of Computing

CISC 271
Study Guide

Complete notes and worked problems for every topic in the course — eigenvalues, SVD, PCA, regression, classification, and more. Built directly from Professor Ellis's course notes.

8
Topics
30+
Worked Problems
4
Exams Covered
How to use this guide Each topic page has: full theory with LaTeX formulas, conceptual explanations, and practice questions with fully worked answers. Click any card below to get started.

01

Course Topics

CLASSES 2 · 7 · 8

Eigenvalues & Eigenvectors

Characteristic polynomials, trace and determinant relations, diagonalization, spectral decomposition, positive definite matrices, quadratic forms, and covariance.

CLASS 3

Graphs & Laplacian Matrix

Graph definitions, adjacency matrix, degree matrix, Laplacian $L = D - A$, eigenvalues of $L$, number of components, Fiedler vector, and spectral clustering.

CLASSES 4 · 5

Vector Spaces & Bases

Column space, null space, RREF, rank-nullity theorem, linear independence, orthogonal bases, orthonormal bases, and orthogonal complement subspaces.

CLASSES 10 · 11 · 12 · 13

Projection & Linear Regression

Standardized data, orthogonal projection to 1D and subspaces, normal equations, residual error, RMS, Hooke's Law, regression with/without intercept, and cross-validation.

CLASSES 14 · 16

Singular Value Decomposition

Derivation from transpose products, left/right singular vectors, matrix spaces from SVD, low-rank approximation (Eckart-Young), polar decomposition, SVD properties.

CLASSES 17 · 19

PCA & Dimensionality Reduction

Zero-mean data, scatter matrix, covariance eigenvalues, PCA as SVD, loading vectors, scores, explained variance, scree plot, and dimensionality reduction.

CLASSES 20 · 21 · 22 · 23 · 25

Classification & Clustering

k-means algorithm, hyperplane from centroids, unit normal form, signed distance, logistic probability, confusion matrix, sensitivity/specificity, ROC curve, AUC.

CLASSES 26 · 28 · 29 · 30 · 31

Artificial Neuron & Kernel Methods

Feed-forward computation, logistic activation, residual error, backpropagation term, gradient descent, weight update, Gram matrix, kernel functions (linear, quadratic, Gaussian).

COURSE SUMMARY · CLASS 34

Complete Reference Sheet

Every formula from the course on one page. Eigenvalues, SVD, PCA, projection, classification, neuron — all key equations for fast last-minute review before the exam.

Exam Analysis

2025 Final Exam Breakdown

April 10, 2025 · 14 questions · 3 hours · Casio FX-991 only

Q1 · 6pts
Eigenvalue perturbation — lower triangular 2×2 Both eigenvalues change (diagonal entries = λ)
Q2 · 6pts
Spectral decomposition — columns of P for symmetric A Columns are orthogonal (symmetric → ortho eigenvectors)
Q3 · 3pts
Definition of projection into a vector space Find weights that minimize Euclidean distance to c
Q4 · 6pts
Projection error — 4×2 matrix Solve normal equation, compute e = c − Aw
Q5 · 6pts
Hooke's Law regression — no intercept k = aᵀc / aᵀa
Q6 · 3pts
SVD null space — last columns of V Null space = right singular vectors with σ = 0
Q7 · 6pts
SVD — find U and V from transpose products Eigenvectors of AAᵀ → U, AᵀA → V
Q8 · 3pts
SVD properties P8.1 / P8.2 / P8.3 P8.1 false (AAᵀ has extra zero eigenvalues)
Q9 · 6pts
PCA first score from SVD of M z₁ = σ₁u₁
Q10 · 6pts
PCA explained variance — effective dimension Cumulative σ²ⱼ / Σσ²ⱼ ≥ θ
Q11 · 6pts
Hyperplane bias from two centroids m = g₁−g₂, h = midpoint, c = −hᵀm/‖m‖
Q12 · 6pts
Logistic probability from (m, b) hyperplane Normalize to unit form, compute p = σ(d)
Q13 · 6pts
Confusion matrix with threshold θ = 2 Classify each point, count TP/FP/TN/FN
Q14 · 3pts
Neuron residual errors with logistic activation r_i = y_i − φ(xᵢw)

Topic Distribution

CLASSIFICATION
~33%
PROJECTION/REGRESSION
~24%
SVD
~19%
PCA
~19%
EIGENVALUES
~19%

Key Takeaways

Triangular matrix eigenvalue trick appears on every test — diagonal entries are the eigenvalues. Know this cold.
Projection/regression questions (Q3–Q5) are very similar to HW4 and Test 2. Practice the normal equation by hand.
SVD null space question reappears in both Test 3 and the final. The null space = last columns of V (zero σ rows).
PCA score formula $\vec{z}_1 = \sigma_1 \vec{u}_1$ is tested directly — it comes from the SVD of the zero-mean data M.
Logistic probability: always convert to unit normal first ($\vec{n} = \vec{m}/\|\vec{m}\|$, $c = b/\|\vec{m}\|$), then apply sigmoid.
Kernel PCA (HW11) did NOT appear on the 2025 final but is in the syllabus — expect it on the 2026 final.