Index Ai Academy
Foundation · Vectors and matrices

SVD and PCA: the important directions first

A matrix as pieces by size, the best rank-k copy, compression, and PCA maps of data.

About 45 minutes · Free lesson

What it covers6

Why it matters

Lesson 61 showed that a matrix of rank r is a sum of r column-times-row pieces. The singular value decomposition picks those pieces in the best possible way: in order of importance, each with a size, its singular value. Keep the biggest few and drop the rest and you have the best small summary of the whole table that any method can give. That one idea compresses pictures, removes noise, and, applied to a data set with its average taken away, becomes PCA, the standard way to squeeze many features into two or three numbers you can plot. Lesson 62's main direction of a cloud is exactly PCA's first piece.

What you'll be able to do

  • Read a matrix as a sum of rank-1 pieces, ordered by size: the singular value decomposition
  • Rebuild a matrix from its first k pieces and watch the error fall as k rises
  • Use np.linalg.svd and torch.linalg.svd, plot the sizes, and pick how many pieces to keep
  • Compress a picture and map 25-number signs down to 2 numbers with PCA
Quick check
Have a guess before you start. A 1,000 × 1,000 picture is stored as 20 column-times-row pieces instead of all its pixels. Roughly how many numbers is that?
  1. About 1,000,000
  2. About 2,000
  3. About 40,000
  4. About 20
Show the answer

About 40,000

Each piece is a column of 1,000 and a row of 1,000 (plus one size): 20 × 2,001 ≈ 40,000, against 1,000,000 pixels.

The math: a matrix as pieces, biggest first

Purpose. The best short summary of a table of numbers.

Every matrix can be written as a sum of rank-1 pieces, each a size times a column times a row:

A=s1u1v1T+s2u2v2T+⋯s1≥s2≥⋯≥0A = s_1 \mathbf{u}_1 \mathbf{v}_1^{\mathsf{T}} + s_2 \mathbf{u}_2 \mathbf{v}_2^{\mathsf{T}} + \cdots \qquad s_1 \ge s_2 \ge \cdots \ge 0

In plain words: each piece is an outer product (lesson 61), scaled by its size. The sizes come biggest first.

The sizes s₁, s₂, … are the singular values. The columns u and the rows v each have length 1 and are perpendicular to one another, the orthonormal directions of lesson 52, so the pieces never overlap: each adds something the others lack.

Quick check
From lesson 61: what is the rank of a sum of two rank-1 pieces whose columns point in different directions?
  1. 2
  2. 0
  3. 1
  4. 4
Show the answer

2

Each piece adds one independent direction, so two different ones give rank 2.

The picture. Watch the arrow sign from lesson 53 rebuilt from its three pieces, sizes 2.449, 1.414 and 1. The picture shades each cell by its number, darker for bigger, so the table reads as the sign it stores. A cell with a thick outline is a pixel that would still round the wrong way at 0.5: on the empty start, the outlines trace the whole arrow. Watch how many outlines each piece clears.

Try it now

The sizes are the arrow's singular values. A photo has many pieces, and most of them are tiny.

Symbols and formula. Stacking the columns u into U, the sizes into a diagonal Σ (sigma) and the rows v into Vᵀ gives the singular value decomposition:

A=UΣVTA = U \Sigma V^{\mathsf{T}}

In plain words: U holds the columns, Σ the sizes on its diagonal, and Vᵀ the rows. Multiplying them out adds up all the pieces.

Keeping the first k. Drop every piece after the k-th and you get A_k, a matrix of rank k. A famous result says no other rank-k matrix is closer to A. The error left behind is set by the sizes you dropped: square them, add, and take the root. Drop only tiny pieces and you lose almost nothing. For the arrow, keeping two pieces drops only the size 1, so the error is 1. Keeping one drops 1.414 and 1, whose squares are 2 (1.414 is √2) and 1, so the error is √3 = 1.732.

Quick check
A matrix has singular values 10, 4, 0.1 and 0.05. How many pieces would you keep for a close copy?
  1. 1: only the biggest matters
  2. 2: the last two are tiny next to 10 and 4
  3. 0: the sizes are too different
  4. 4: every piece matters equally
Show the answer

2: the last two are tiny next to 10 and 4

10 and 4 carry almost everything; 0.1 and 0.05 add very little. Two pieces give a close, rank-2 copy.

In your own words
In your own words: why is keeping the biggest pieces the best way to shrink a matrix?

Write a sentence or two, then compare it with a good answer in the lesson.

In Python: np.linalg.svd

Purpose. Get every piece at once, then rebuild from the first k.

import numpy as np

A = np.array([[4, 2, 4], [2, -2, -1]])
U, s, Vt = np.linalg.svd(A)
print(np.round(s, 6).tolist())
A1 = s[0] * np.outer(U[:, 0], Vt[0])
print(np.round(A1, 6).tolist(), round(np.linalg.norm(A - A1), 6))
It runs right here in your browser, with nothing to install.

np.linalg.svd returns three things: U, whose columns are the u's; s, the sizes, biggest first; and Vt, whose rows are the v's. Piece k is s[k] * np.outer(U[:, k], Vt[k]). np.linalg.norm of the difference measures the error, here 3.

Code flow. Step through the rebuild, one piece per pass of the loop. To keep every number exact, the code types in the two pieces the SVD found: sizes 6 and 3, columns (1, 0) and (0, 1), and rows (2, 1, 2) ÷ 3 and (2, −2, −1) ÷ 3. (Your computer may hand back the first column and row with both signs flipped. Flipping both leaves the piece the same.) Watch Ak gain row 1 on the first pass and row 2 on the second.

The first k pieces in one line. Slices keep the first k columns of U, sizes and rows of Vt:

import numpy as np

arrow = np.array([[0,0,1,0,0],[0,1,1,1,0],[1,0,1,0,1],[0,0,1,0,0],[0,0,1,0,0]])
U, s, Vt = np.linalg.svd(arrow)
print(np.round(s, 3).tolist())
for k in [1, 2, 3]:
    Ak = U[:, :k] @ np.diag(s[:k]) @ Vt[:k]
    print(k, round(np.linalg.norm(arrow - Ak), 3))

The arrow sign's sizes are 2.449, 1.414, 1 and then zeros: its rank is 3. The error falls as k rises and hits 0 at k = 3. A named mistake: Vt is already transposed. Its rows are the v's, so the k-th is Vt[k], not Vt[:, k].

The shapes of the three pieces. For an m × n matrix, U is m × m, s holds as many sizes as the smaller of m and n, and Vt is n × n. Only the first few columns of U and rows of Vt matter when you keep only a few pieces, which is why rebuilding uses the slices U[:, :k] and Vt[:k]. A 1,000 × 1,000 picture kept at k = 20 needs just the first 20 columns of U, 20 sizes, and the first 20 rows of Vt.

Quick check
What does this print?
import numpy as np

U, s, Vt = np.linalg.svd(np.array([[2, 0], [0, 1], [0, 0]]))
print(np.round(s, 6).tolist(), U.shape, Vt.shape)
  1. [2.0, 1.0] (3, 2) (2, 3)
  2. [2.0, 1.0] (3, 3) (2, 2)
  3. [1.0, 2.0] (3, 3) (2, 2)
  4. [2.0, 1.0, 0.0] (3, 3) (2, 2)
Show the answer

[2.0, 1.0] (3, 3) (2, 2)

A 3 × 2 matrix has 2 sizes (the smaller of 3 and 2), biggest first. U is 3 × 3 and Vt is 2 × 2.

Worked example

Worked example: rank 1, rank 2 and full rebuilds

Rebuild the arrow from 1, 2 and 5 pieces, and test whether each is still readable: round each pixel to 0 or 1 at 0.5 and compare with the original.

import numpy as np

arrow = np.array([[0,0,1,0,0],[0,1,1,1,0],[1,0,1,0,1],[0,0,1,0,0],[0,0,1,0,0]])
U, s, Vt = np.linalg.svd(arrow)
for k in [1, 2, 5]:
    Ak = U[:, :k] @ np.diag(s[:k]) @ Vt[:k]
    readable = np.array_equal(Ak > 0.5, arrow == 1)
    print(k, round(np.linalg.norm(arrow - Ak), 3), readable)
In the lesson you can change this and run it.

One piece gives a blur, error 1.732, and some pixels land on the wrong side of 0.5. Two pieces are already readable: every pixel rounds the right way, though the error is still 1. Five pieces rebuild the sign exactly. Ak > 0.5 gives True wherever a pixel is inked after rounding, and arrow == 1 does the same for the original, so np.array_equal compares the two pictures pixel by pixel.

Quick check
What does this print?
import numpy as np

A = np.array([[3, 0], [0, 1]])
U, s, Vt = np.linalg.svd(A)
print(np.round(s, 6).tolist(), A.shape)
  1. [1.0, 3.0] (2, 2)
  2. [3.0, 1.0] (2, 2)
  3. [4.0] (2, 2)
  4. [3.0, 0.0] (2, 2)
Show the answer

[3.0, 1.0] (2, 2)

For a diagonal matrix with positive entries the sizes are its diagonal, biggest first: 3 then 1.

Plot the sizes. Drawing the sizes in order shows at a glance how many pieces matter. Here are the arrow's:

import numpy as np
import matplotlib.pyplot as plt

arrow = np.array([[0,0,1,0,0],[0,1,1,1,0],[1,0,1,0,1],[0,0,1,0,0],[0,0,1,0,0]])
U, s, Vt = np.linalg.svd(arrow)
plt.plot([1, 2, 3, 4, 5], s)
plt.xlabel("piece k")
plt.ylabel("size of piece k")
plt.show()

A photo's plot has the same shape over many more pieces: a steep drop, then a long, low tail of pieces you can drop.

Quick check
Read the plot. Where does the line first reach 0, and what does that say about the arrow?
  1. At piece 5: all five pieces matter, so its rank is 5
  2. Nowhere: the sizes only shrink, so its rank is 5
  3. At piece 4: only three pieces have any size, so its rank is 3
  4. At piece 3: only two pieces have any size, so its rank is 2
Show the answer

At piece 4: only three pieces have any size, so its rank is 3

The sizes are 2.449, 1.414 and 1, then 0 and 0. The line reaches 0 at piece 4, so three pieces carry the whole arrow: its rank is 3, and three pieces rebuild it exactly.

Drill

Rebuild from k pieces

Fix rebuild(A, k) so it returns the sum of the first k pieces. Then set errs to the rebuild error for k = 1, 2 and 3 on the cross sign, each rounded to 3 places. .item() turns NumPy's number into a plain Python one, so the list prints as plain numbers.

import numpy as np

def rebuild(A, k):
    U, s, Vt = np.linalg.svd(A)
    return U[:, :k] @ np.diag(s[:k]) @ Vt[:, :k]

cross = np.array([[1,0,0,0,1],[0,1,0,1,0],[0,0,1,0,0],[0,1,0,1,0],[1,0,0,0,1]])
errs = []
print(errs)
In the lesson, your code is checked as you write it.
In your own words
Why can a 5 × 5 sign need only 2 or 3 pieces, while a photo needs many?

Write a sentence or two, then compare it with a good answer in the lesson.

In AI and ML: compression and maps of data

Purpose. Squeeze many numbers into a few while keeping what matters.

The pieces do two jobs in machine learning. They compress a matrix, whether it holds a picture or a layer's weights, and they map data with many features down to a few numbers you can plot.

Compression. The smallest k that keeps a sign readable depends on the sign: the box needs 1 piece, the arrow 2, and the cross 3. A photo works the same way with bigger numbers: a few dozen pieces often look almost like the original. Weight matrices too: when a layer's sizes fall fast, a few pieces can stand in for the whole layer, the idea behind lesson 61's low-rank updates.

In PyTorch. You met torch.linalg.svd in lesson 52, where the rows of its Vh passed the orthonormal test. It hands back the same three things as NumPy, named U, S and Vh (Vt by another name). Here is the arrow kept at two pieces, in the 64-bit decimals NumPy uses:

import torch

arrow = torch.tensor([[0,0,1,0,0],[0,1,1,1,0],[1,0,1,0,1],[0,0,1,0,0],[0,0,1,0,0]], dtype=torch.float64)
U, S, Vh = torch.linalg.svd(arrow)
print(S.round(decimals=3).tolist())
A2 = U[:, :2] @ torch.diag(S[:2]) @ Vh[:2]
print(round(torch.linalg.norm(arrow - A2).item(), 3), torch.equal(A2 > 0.5, arrow == 1))

The sizes are the ones NumPy printed, 2.449, 1.414, 1 and two zeros, and two pieces leave the same error, 1.0. torch.equal asks whether two tensors match everywhere: True, so two pieces keep the arrow readable.

PCA: principal component analysis. Take a data matrix X (one row per example), subtract the average row from every row so the cloud is centered on the origin, and take the SVD. The rows of Vt are the principal directions, the first being lesson 62's main direction of spread. Each example's position along the first two gives it two numbers, a point you can plot:

coordinates=Xc V2V2=the first two principal directions, as columns\text{coordinates} = X_c \, V_2 \qquad V_2 = \text{the first two principal directions, as columns}

Here are nine signs, three of each type, some with one pixel flipped, each 25 numbers long, mapped to 2 numbers:

import numpy as np

cross = np.array([[1,0,0,0,1],[0,1,0,1,0],[0,0,1,0,0],[0,1,0,1,0],[1,0,0,0,1]])
arrow = np.array([[0,0,1,0,0],[0,1,1,1,0],[1,0,1,0,1],[0,0,1,0,0],[0,0,1,0,0]])
box = np.array([[1,1,1,1,1],[1,0,0,0,1],[1,0,0,0,1],[1,0,0,0,1],[1,1,1,1,1]])
rows = []
for S in [cross, arrow, box]:
    rows.append(S.flatten())
    for spot in [(0, 0), (2, 2)]:
        T = S.copy()
        T[spot[0], spot[1]] = 1 - T[spot[0], spot[1]]
        rows.append(T.flatten())
X = np.array(rows, dtype=float)
mean = np.zeros(25)
for i in range(9):
    mean = mean + X[i] / 9
Xc = X.copy()
for i in range(9):
    Xc[i] = X[i] - mean
U, s, Vt = np.linalg.svd(Xc)
coords = Xc @ Vt[:2].T
print(np.round(coords, 2).tolist())
share = (s[0] ** 2 + s[1] ** 2) / (s @ s)
print(np.round(s[:4], 3).tolist(), round(share, 3))

dtype=float makes the array hold decimals, so the centered values aren't cut to whole numbers. The nine rows land in three tight groups of three, one group per sign type, though each sign started as 25 numbers. (Which way each axis points can flip from one computer to another; the groups don't change.) A size squared measures the spread along its direction, so share is the part of all the spread the first two directions keep: 0.915.

Quick check
Why is the average row subtracted from every row before the SVD in PCA?
  1. To remove the redundant features first
  2. To make every number positive
  3. To make the matrix square
  4. So directions measure spread around the center
Show the answer

So directions measure spread around the center

Without centering, the first direction would mostly point at the average itself. Centering puts the cloud at the origin, so the directions measure how examples differ.

Drill

The smallest readable k

Write smallest_k(S) returning the smallest k for which the rebuild from k pieces, rounded at 0.5, matches the sign exactly. Then set ks to its answers for the box, the arrow and the cross, in that order.

import numpy as np

def rebuild(A, k):
    U, s, Vt = np.linalg.svd(A)
    return U[:, :k] @ np.diag(s[:k]) @ Vt[:k]

def smallest_k(S):
    return 5

cross = np.array([[1,0,0,0,1],[0,1,0,1,0],[0,0,1,0,0],[0,1,0,1,0],[1,0,0,0,1]])
arrow = np.array([[0,0,1,0,0],[0,1,1,1,0],[1,0,1,0,1],[0,0,1,0,0],[0,0,1,0,0]])
box = np.array([[1,1,1,1,1],[1,0,0,0,1],[1,0,0,0,1],[1,0,0,0,1],[1,1,1,1,1]])
ks = []
print(ks)
In the lesson, your code is checked as you write it.
In your own words
Nine 25-number signs became nine points on a plane, and the three types separated. What did PCA keep, and what did it throw away?

Write a sentence or two, then compare it with a good answer in the lesson.

Goes deeper in

Practice what you just read

  • 6 code examples you can edit and run
  • a practice exercise that checks your code
  • a 5-question quiz
  • 9 review cards for spaced repetition

Try a quiz question

What do the singular values of a matrix tell you?

  1. How big each rank-1 piece is, biggest first
  2. The matrix's entries in sorted order
  3. The eigenvalues of every row
  4. How many rows the matrix has

Answer it at the end of the lesson to see if you're right, and why.

Before this lesson

Keep learning

Where this lesson leads.

575 lessons in thirteen phases. 176 are free without an account, and a free account opens the rest. See every lesson.

Foundation · about 96 hours of study

Foundation math

From whole numbers to the math inside a neural network: each idea by hand, in Python, and in AI 118 lessons All free
  1. Numbers and place value FreeWhat a digit's place is worth, the number line, and Python as a calculator that never tires.36 min
  2. Multiplying and dividing: groups FreeEqual groups, quotient and remainder, and cutting training data into batches.35 min
  3. Order of operations, and names for numbers FreeBrackets, powers, then multiply and divide, then add and subtract; variables hold numbers.36 min
  4. Numbers below zero FreeThe number line through zero, sign rules, distance from zero, and an error's direction and size.37 min
  5. Fractions: equal parts, equal fractions, adding FreePieces of a whole, equal fractions, and re-cutting to add; accuracy as a fraction.45 min
  6. Multiplying and dividing fractions FreeA share of a share, dividing by flipping, and why chances in a row get small.35 min
  7. Decimals and rounding FreeFractions in place value, rounding, and why 0.1 + 0.2 isn't exactly 0.3.36 min
  8. Percentages FreePer hundred, percent change, percentage points, and the accuracy and error rates of models.35 min
  9. Ratios, rates and proportion FreeSharing in a ratio, rates per unit, scaling, and splitting data 80/10/10.35 min
  10. Powers and roots FreeRepeated multiplying, the laws of exponents, square roots, and squared error.43 min
  11. Very big and very small numbers FreePowers of ten, e-notation like 3e-4, orders of magnitude, and a model's size in memory.36 min
  12. Bits, bytes and floats: how numbers are stored FreeBase 2, what n bits can hold, a float as sign, power and digits, and a tensor's memory.35 min
  13. Review: find the slip (numbers) FreeBroken calculations to fix, mixed practice from the whole module, and its formula sheet.34 min
  14. Letters for numbers FreeVariables, terms and like terms, substitution, and a model's formula as a Python function.41 min
  15. Expanding and factoring FreeBrackets as areas, pulling out common factors, and a squared error expanded into a bowl.41 min
  16. Solving equations: keep the balance FreeThe same move on both sides, undoing steps in reverse, and running a model's formula backward.41 min
  17. Rearranging a formula FreeMaking any letter the subject, round-trip checks, and planning a training run.36 min
  18. Inequalities and true-or-false FreeRanges, flipping the sign, if and elif, and the thresholds, clips and ReLU inside a model.42 min
  19. Functions: a rule from input to output Freef(x) notation, tables and graphs, domain and range, loops over lists, and loss as a function.42 min
  20. Chaining and undoing functions Freeg(f(x)), inverses that undo, scaling and unscaling, and a network as a chain.42 min
  21. Straight lines: slope and intercept FreeRise over run, the line through two points, and a model's weight and bias as slope and intercept.42 min
  22. Two equations, two unknowns FreeSubstitution and elimination, where lines cross, and why models are fitted, not solved.42 min
  23. Quadratics: bowls and their lowest point FreeRoots by factoring and by formula, the discriminant, the vertex, and the best weight of a model.42 min
  24. Exponentials: growth and decay FreeThe same factor each step, the number e, and decaying learning rates.43 min
  25. Logarithms: the power question FreeTo what power?, the laws of logs, and adding log-probabilities instead of multiplying.45 min
  26. Sequences and Σ FreeArithmetic and geometric sequences, Σ as a loop that adds, and mean squared error.40 min
  27. Weighted and moving averages FreeWeights that add to 1, a window average, the exponential moving average, and smoothing a loss curve.35 min
  28. Sets, and reading the symbols FreeSets, union and intersection, ℝⁿ and f : ℝⁿ → ℝᵐ, and reading a paper's first equation.41 min
  29. Review: find the slip (algebra) FreeBroken algebra and code to fix, mixed practice from the whole module, and its formula sheet.34 min
  30. Transposing step by step: undo in reverse order FreeA formula's layers, the opposite of each, peeling from the outside in, conditions, and round-trip checks.35 min
  31. Formulas with fractions: clear them first FreeClearing denominators, cross-multiplying, a letter in the bottom, and reciprocal formulas like 1/R = 1/R₁ + 1/R₂.35 min
  32. When the letter appears twice: collect and factor FreeGathering the letter on one side, factoring it out, dividing by the bracket, and rearranging precision and F1.35 min
  33. Squares, roots and powers FreeSquare roots with ±, squaring to undo a root, cube and nth roots, and the √d inside attention.40 min
  34. When the letter is squared and plain: quadratics FreeRearranging into ax² + bx + c = 0, the formula, completing the square, and choosing the root that fits.35 min
  35. Letters in the power: exponentials and logarithms FreeTaking logs to bring a power down, exponentiating to free a letter, and the logit that undoes the sigmoid.40 min
  36. Rearranging inequalities: when the sign flips FreeThe four moves on an inequality, negatives and reciprocals that flip it, logs below 1, and sample sizes.35 min
  37. Inverse functions, lines and scaling FreeUndoing a function by making x the subject, domains that allow it, line forms, and unscaling predictions.40 min
  38. Sequences and sums: find n, a, d and r FreeEvery letter of the arithmetic and geometric formulas as the subject, and an average's window from its decay.40 min
  39. Combining formulas: substitute and eliminate FreePutting one formula inside another to remove a letter, and sizing a training run from a time or money budget.35 min
  40. Review: rearrange anything (algebra) FreeBroken rearrangements to fix, mixed practice from the whole algebra block, a strategy chart, and its formula sheet.35 min
  41. Coordinates and distance FreePoints as (x, y), Pythagoras, distance in any number of coordinates, and nearest neighbors.42 min
  42. Angles, degrees and radians FreeMeasuring a turn, converting units, arcs, and the angle between two arrows.36 min
  43. Sine and cosine FreeAcross and up: right triangles, the unit circle, and directions and rotations.42 min
  44. Waves: height, speed and shift FreeAmplitude, period and phase, sampling a wave, and position codes in transformers.37 min
  45. Review: find the slip (shapes and waves) FreeBroken geometry and code to fix, mixed practice from the module, and its formula sheet.29 min
  46. Angles, triangles and waves: rearranging with sin, cos and tan FreeSides and angles from SOH CAH TOA, principal values, arcs, polar coordinates, and a wave's time and period.45 min
  47. Vectors: lists of numbers that point FreeA list, a point and an arrow; components, dimension, NumPy arrays, and data as vectors.42 min
  48. Adding and scaling vectors FreeComponent-wise sums, scalar multiples, mixes, and the weight update w − ηg.43 min
  49. Length and distance: norms FreeL2, L1 and max norms, unit vectors, distance, and the norms inside losses and clipping.42 min
  50. The dot product FreeMultiply in pairs and add; a·a as squared length; a @ b; one neuron as a dot product.43 min
  51. Angle, cosine similarity and projection FreeDirection without length, perpendicular vectors, shadows, and search by meaning.45 min
  52. Perpendicular directions: bases and coordinates FreeOrthonormal directions, coordinates as dot products, rebuilding a vector, and taking a direction out.35 min
  53. Matrices: tables of numbers FreeRows, columns and entries; two-dimensional arrays and slices; a data set as a matrix.41 min
  54. Adding, scaling and transposing matrices FreeEntry-by-entry sums and scalings, the transpose, mean images, and tables turned around.36 min
  55. A matrix times a vector FreeEach row dotted with the vector, the size rule, A @ x, and a layer acting on one example.42 min
  56. Matrix multiplication FreeRow meets column, the size rule, order matters, and a batch through a layer: X W + b.45 min
  57. Identity, diagonal, triangular and symmetric FreeFour patterns and their jobs: rescaling, distance tables, and the causal mask.37 min
  58. Matrices as movements of space FreeStretches, flips, shears and rotations; columns as landing spots; chaining; a layer moving data.42 min
  59. Determinant and inverse FreeThe area factor ad − bc, singular matrices, undoing a matrix, and redundant features.42 min
  60. Solving systems by elimination FreeAugmented matrices, row moves, back-substitution, the three outcomes, and fitting weights.45 min
  61. Independence, span and rank FreeRedundant vectors, span, rank, outer products, and low-rank updates like LoRA.42 min
  62. Eigenvectors: directions that only stretch FreeAv = λv, checking candidates, np.linalg.eig, and the main direction of a data cloud.41 min
  63. SVD and PCA: the important directions first FreeA matrix as pieces by size, the best rank-k copy, compression, and PCA maps of data.45 min
  64. Tensors, shapes and broadcasting FreeAxes and shapes, reshape, axis=, broadcasting, and shapes through a small model.43 min
  65. One-hot vectors and lookup tables: embeddings FreeA one-hot row picks a table's row, a sentence looked up as vectors, and a score for every word.35 min
  66. Sliding dot products: convolution FreeA small pattern slid along a list and over a grid, output sizes, stride and padding, and shared weights.40 min
  67. Review: find the slip (vectors and matrices) FreeShape slips to fix, mixed practice from the module, and its formula sheet with shapes.36 min
  68. Vector equations: rearranging with arrows FreeSolving for a vector or a scale, recovering a gradient from two weight snapshots, and an angle from a dot product.40 min
  69. Matrix equations: multiply by the inverse, on the correct side FreeNo dividing by a matrix: inverses from the left or right, order, transposes, and the normal equations.40 min
  70. Rates of change FreeChange over change, secants, rates from readings with np.diff, and reading a loss curve.41 min
  71. Limits: what a value heads toward FreeApproaching from both sides, holes, corners and jumps, epsilon, and judging activations.37 min
  72. The derivative: slope at a point FreeShrinking secants, f′(x) and dy/dx, Δy ≈ f′Δx, and which way a weight should move.45 min
  73. Derivative rules FreeConstants, powers, multiples, sums, eˣ and ln x, checked numerically, and a squared error's slope.41 min
  74. The product rule and the chain rule FreeProducts and chains of functions, forward then backward, and one neuron's dL/dw.45 min
  75. Lowest points and curvature Freef′ = 0, the second-derivative test, tables, local and global minima, and convex losses.43 min
  76. Standing in for a curve near a point FreeTangent lines, a curvature term, growing error, and why learning rates must be small.37 min
  77. Partial derivatives FreeOne input at a time, ∂, nudging in Python, and the two slopes of a line's loss.41 min
  78. The gradient FreeAll the slopes as a vector, steepest climb and descent, a numerical gradient, and the loss gradient.42 min
  79. Gradient descent FreeStepping against the gradient, the learning rate, a training loop, and fitting a line.45 min
  80. Faster descent: momentum and Adam FreeMomentum and Adam by hand, the bias correction, and three optimizers down one narrow valley.40 min
  81. The chain rule with many paths: backpropagation by hand FreeGraphs, multiply along paths and add across them, gradient checks, and a 2-2-1 network by hand.45 min
  82. Automatic differentiation: how PyTorch finds every slope FreeA tape of steps walked backward, and requires_grad, backward(), .grad, zero_grad() and no_grad().40 min
  83. Jacobians and Hessians FreeSlopes of vector functions, a layer's Jacobian, second partials, and bowls, domes and saddles.42 min
  84. Integrals: adding up thin slices FreeStrip and trapezoid sums, summing undoes differencing, probability as area, and AUC.43 min
  85. Review: find the slip (calculus) FreeCalculus slips to fix, mixed practice from the module, and its formula sheet.36 min
  86. Rearranging in calculus: slopes, steps and constants FreeSetting a slope to zero, Newton's step, the chain rule turned around, constants from a point, and learning rates.45 min
  87. Counting: choices, orders and groups FreeThe counting principle, factorials, ordered picks and groups, and why attention's cost grows with the square of the length.43 min
  88. Probability: from counts to chances FreeEqually likely outcomes, the complement, distributions, seeded simulation, and a next-word model.43 min
  89. Combining events: and, or, independent FreeThe or rule, independence, probability trees, drawing with and without replacement, and chains of steps.42 min
  90. Conditional probability and Bayes' rule FreeConditionals from tables, why P(A | B) isn't P(B | A), Bayes' rule, and precision and recall.45 min
  91. Random variables and distributions FreeUncertain numbers, the probability of a range, sampling by slices, and sampling versus argmax.42 min
  92. Expected value, variance and standard deviation FreeProbability-weighted averages, spread in the data's own units, and a training loss as an expected value.42 min
  93. Bernoulli, binomial, uniform and normal FreeYes-or-no trials, counts of yeses, the 68-95-99.7 rule, z-scores, and an accuracy's error bar.45 min
  94. Two variables at once: covariance and correlation FreeJoint tables and marginals, covariance, correlation, the covariance matrix, and features that carry the same information.45 min
  95. Review: find the slip (probability) FreeProbability slips to fix, mixed practice from the module, and its formula sheet.36 min
  96. Probability formulas: solve for the missing chance FreeConditionals, Bayes' rule for the prior, the or rule, expected values, variance, and z-scores turned around.45 min
  97. Describing data: center, spread and scale FreeMean, median and mode, variance and standard deviation, percentiles, outliers, and standardizing features.42 min
  98. Samples, the law of large numbers and the central limit theorem FreeSamples and populations, the standard error σ/√n, the bell of averages, and error bars on test scores.43 min
  99. Is the difference real? Comparing two models FreePaired differences, a bootstrap interval, what a p-value is and isn't, and why many tries make lucky wins.40 min
  100. Maximum likelihood FreeLikelihoods as products, log-likelihoods as sums, grid search, underflow, and the negative log-likelihood loss.45 min
  101. Linear regression: the best line FreeResiduals, least squares by formula, three roads to one line, the matrix form, and train versus test error.45 min
  102. Logistic regression: a line that answers yes or no FreeThe sigmoid, the log loss, its slope p − y, a training loop, and choosing a threshold.45 min
  103. Scoring a classifier: confusion table, precision, recall, F1 FreeThe four cells, accuracy, precision, recall and F1, and what moving a threshold does to each.35 min
  104. Softmax: scores into probabilities FreeExponentiate and share out, shifting and temperature, the safe softmax, and a classifier's last step.43 min
  105. Surprise, entropy and cross-entropy FreeSurprise in bits, entropy, cross-entropy and KL divergence, perplexity, and a language model's loss.45 min
  106. KL divergence: how far one distribution is from another FreeKL by hand, cross-entropy as entropy plus KL, why it isn't symmetric, and a soft-target loss.35 min
  107. Review: find the slip (statistics and information) FreeStatistics slips to fix, mixed practice from the module, and its formula sheet.38 min
  108. Statistics and information: n, z, logits and bits FreeSample sizes from error bars, inputs from probabilities, logit gaps from softmax, and chances from bits and losses.40 min
  109. One neuron, start to finish FreeWeighted sum, activation, loss, slopes and update, by hand and in code, and what one straight line can and can't split.45 min
  110. A layer and a small network as matrices FreeTwo layers on a batch, the shape of every matrix, adding a hidden unit, and XOR solved.45 min
  111. Training a small network FreeThe backward pass in matrix form, gradient checks, a full training loop, and reading loss curves.45 min
  112. The same network in PyTorch Freenn.Sequential with the NumPy weights loaded, the same loss and slopes, then SGD and mini-batches.45 min
  113. Attention as matrix math FreeScaled dot-product scores, the causal mask, a row-by-row softmax, and blending value vectors.45 min
  114. Residual connections and layer norm FreeAdding the input back and why its slope keeps learning alive, layer norm by hand, and add and norm.40 min
  115. A transformer block, start to finish FreeA sentence through embeddings, attention, add and norm, a small network and next-word scores, every shape named.45 min
  116. Planning a model: rearranging the formulas of a working network FreeLayer widths from parameter counts, sizes from memory and compute, the attention scale, and a training plan.40 min
  117. Final test: transpose any formula FreeSlips to fix from every topic, mixed practice of every kind of rearranging, the strategy chart and the full sheet.45 min
  118. Capstone: a sign reader from the formulas FreeBuild, train and test a trail-sign classifier with NumPy alone, and report its accuracy with an error bar.45 min
Part of a free course 118 lessons

The Foundation Math For Ai & ML: from arithmetic to attention

Learn the math behind AI from arithmetic up: numbers, algebra, vectors and matrices, calculus, probability and statistics, each shown as numbers moving and checked in Python, ending with a network you build and train from the formulas.

Never trained an AI? Teach a small one to write, right in your browser. It takes about 20 seconds and needs no account. Teach it to write