Advanced Topics Preview
Beyond matrices: tensors, randomized algorithms, quantum computing, topological data analysis, and category theory — five frontiers built on the linear algebra you now know.
Matrices Are 2-Tensors
A scalar is a 0-tensor. A vector is a 1-tensor. A matrix is a 2-tensor. A 3-tensor T ∈ ℝ^{I×J×K} is a three-dimensional array — stack of matrices. Neural network weights, video frames, and medical scans are all tensors.
Tucker = Tensor SVD
Tucker decomposition generalizes SVD: T ≈ G ×₁ A ×₂ B ×₃ C where G is a small core tensor and A, B, C are orthonormal factor matrices. Used in multi-way PCA, neuroscience, and chemometrics. Tensor rank is NP-hard to compute — unlike matrix rank.
Sketch First, Compute Less
Exact SVD costs O(nd²). Randomized SVD: multiply by a random matrix Ω to sketch Y = AΩ, orthonormalize Q, then compute the small matrix B = QᵀA. Near-optimal approximation with high probability.
High Dimensions, Low Projection
Any n points in ℝᵈ can be projected into k = O(log n / ε²) dimensions while preserving all pairwise distances to within 1 ± ε. The target dimension depends only on n and ε — not on d. Random projections work.
Qubits Are Unit Vectors
A qubit |ψ⟩ = α|0⟩ + β|1⟩ is a unit vector in ℂ². Gates are unitary matrices. Measurement is projection onto an eigenspace. An n-qubit system lives in ℂ^{2ⁿ} — an exponentially large space.
Tensor Products Entangle
Two-qubit states live in ℂ² ⊗ ℂ². If the state cannot be written as a product, the qubits are entangled. Entanglement is a linear algebra statement: the two-qubit state matrix has rank > 1. Quantum speedup exploits constructive and destructive interference.
Count the Holes
TDA asks: what are the connected components, loops, and voids in a dataset? Homology groups Hₖ = ker(∂ₖ)/im(∂ₖ₊₁) are vector spaces. Betti numbers β₀, β₁, β₂ count components, loops, voids — computed by matrix reduction over 𝔽₂.
Features That Persist Are Real
Sweep ε from 0 to ∞ and track which topological features are born and die. A feature that persists over a wide range is likely real structure; a short-lived feature is noise. The output is a persistence diagram — a topological fingerprint.
Functors Generalize Linear Maps
Categories: objects + morphisms. Vect is the category of vector spaces. A functor maps categories preserving structure. The tensor-Hom adjunction Hom(U⊗V, W) ≅ Hom(U, Hom(V,W)) is why matrices can be viewed as bilinear forms and linear operators simultaneously.
- Transpose: functor Vect → Vectᵒᵖ
- Determinant: functor GL(n) → ℝ*
- Double-dual: natural transformation V → V**
The Frontier Awaits
Every advanced topic is a generalization of what you already know. Tensors ↔ matrices. Random SVD ↔ exact SVD. Qubits ↔ unit vectors. Homology ↔ null space. Functors ↔ linear maps. The foundation you built carries you everywhere.