Index Ai Academy
Foundation · Putting it together

Attention as matrix math

Scaled dot-product scores, the causal mask, a row-by-row softmax, and blending value vectors.

About 45 minutes · Free lesson

What it covers6

Why it matters

Attention is the step that made today's language models possible, and it has a reputation for being hard. It isn't, once its pieces are named, because every one of them is a lesson you've done. Words are vectors (lesson 47), looked up in a table by their ids (lesson 65). How well two words match is a dot product (lessons 50 and 51). The scores are scaled down by √d (lesson 10's square roots) so softmax doesn't saturate. Words aren't allowed to look ahead, which is the triangular mask of lesson 57. The scores become weights with a row-by-row softmax (lesson 104). And each word's new vector is a weighted average of the others (lessons 48 and 92). Put together, that's one line of math, softmax(QKᵀ/√d)V, and eight lines of NumPy. This lesson builds it, then shows what it's for: letting "warm" find "fire".

What you'll be able to do

  • Build attention from pieces you know: dot products for similarity, a 1/√d scale, a causal mask, softmax row by row, and a weighted average
  • Compute one row of attention weights and one output vector by hand for three words
  • Write attention in eight lines of NumPy for five words, with the shape at every step, and add the mask
  • Read what attention does for a language model: each word's vector becomes a blend of the words it attends to, and run it in PyTorch with W_Q, W_K and W_V as nn.Linear layers
Quick check
Have a guess before you start. Two word vectors point in nearly the same direction. What does that suggest about the words, and which operation measures it?
  1. They're opposites; a sum measures it
  2. They're the same word; a norm measures it
  3. They're related; a dot product measures it
  4. Nothing; vectors don't carry meaning
Show the answer

They're related; a dot product measures it

Lesson 51: vectors pointing the same way have a large dot product and cosine near 1. Attention uses exactly this to decide which words go together.

The math: score, scale, mask, softmax, blend

Purpose. Give each word a new vector that blends in the words it's most related to.

Each word has three vectors: a query q (what it's looking for), a key k (what it offers to be found by) and a value v (what it passes on). For one word's query, attention does five things:

  1. Score every word: the dot product q · kⱼ, large when the key matches the query.
  2. Scale each score by 1/√d, where d is the vectors' length, so long vectors don't give huge scores.
  3. Mask the words that come later: their scores become −∞, so they get no weight.
  4. Softmax the scores into weights that are positive and add up to 1.
  5. Blend the values: the output is Σ weightⱼ × vⱼ, a weighted average.

Three words, "the", "fire" and "was", with keys (1, 0), (0, 1) and (1, 1) and values (1, 0), (0, 1) and (0.5, 0.5). The word "fire" asks with the query (1, 2).

Quick check
From lesson 104: what does softmax guarantee about a row of attention weights?
  1. The largest weight is exactly 1
  2. Every weight is the same
  3. All positive, adding up to 1
  4. The weights equal the scores
Show the answer

All positive, adding up to 1

So the output is a true weighted average of the value vectors: a blend, not a sum that grows with the number of words.

The picture. Step through "fire" attending to all three words, first with nothing hidden, the way a model reading a whole sentence at once would. The picture lays the whole calculation out as one table, one row per word: its key, its score, its weight and its value. The query sits on top, over the key columns, so each score is the query dotted with the key in its own row. When softmax runs, the weight column is shaded by each word's share, and the last row adds up the value columns, every value times its word's weight. Then hide a word with the mask: as the second word, "fire" may only look at "the" and itself.

Try it now

Then set the mask to 1, 1, 0 and press Use these numbers: "was" comes after "fire", so its score becomes −∞, its weight drops to 0, and the other two share all of it: 0.33 and 0.67.

Symbols and formula. Stack every word's query, key and value as the rows of Q, K and V (n × d each):

Attention(Q,K,V)=softmax ⁣(QK⊤d+M)V\text{Attention}(Q, K, V) = \text{softmax}\!\left(\frac{Q K^{\top}}{\sqrt{d}} + M\right) V

In plain words: score every query against every key at once, scale, add the mask M (0 where allowed, −∞ where not), softmax each row, and blend the values with those weights.

QKᵀ is n × n: entry (i, j) is how much word i's query matches word j's key. After the softmax, row i holds word i's weights over all words, and multiplying by V (n × d) gives n × d: one new vector per word. For "fire" with nothing hidden: the scores are 1, 2 and 3, scaled by 1/√2 to 0.71, 1.41 and 2.12; softmax gives 0.140, 0.284 and 0.576; and the output is 0.140 × (1, 0) + 0.284 × (0, 1) + 0.576 × (0.5, 0.5) = (0.428, 0.572). With the mask, "was" drops out and the weights become 0.330 and 0.670.

Quick check
Queries and keys have length d = 16. By what is every score divided?
  1. 16
  2. 2
  3. 4
  4. 256
Show the answer

4

The scale is √d = √16 = 4. Without it, scores grow with d and softmax becomes nearly one-hot.

In your own words
In your own words: why is the output a weighted average of the values rather than a plain sum?

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

In Python: eight lines for five words

Purpose. Compute attention for a whole sentence at once, with the causal mask.

import numpy as np

X = np.array([[1.0, 0.0, 0.0, 0.0], [0.0, 1.0, 1.0, 0.0], [0.0, 0.0, 1.0, 0.0],
              [0.0, 1.0, 0.0, 1.0], [1.0, 0.0, 0.0, 1.0]])
d = X.shape[1]
S = X @ X.T / np.sqrt(d)
mask = np.tril(np.ones((5, 5)))
S = np.where(mask == 1, S, -np.inf)
E = np.exp(S - S.max(axis=1, keepdims=True))
A = E / E.sum(axis=1, keepdims=True)
out = A @ X
print(S.shape, A.shape, out.shape)
print(A[3].round(3).tolist())
It runs right here in your browser, with nothing to install.

Five camp-log words, "the fire was warm tonight", each a 4-number vector, are the rows of X; here the queries, keys and values are all X itself. S is 5 × 5, one score per pair; np.where keeps a score where the mask has a 1 and puts −∞ where it has a 0; A's rows are the weights; out is 5 × 4, a new vector for every word. Row 3 is "warm": weights 0.157, 0.259, 0.157 and 0.427 on the first four words, and 0 on "tonight", which comes later.

Code flow. Step through the same lines one at a time and watch each matrix appear as a grid of boxes. Two lines are folded together here: the mask goes straight into the exponential, so the hidden places of E come out as exactly 0, e to −∞. And with scores this small, the flow skips subtracting each row's largest score first. Keep an eye on row 3 of A, "warm".

A named mistake: multiplying the scores by the mask. Lesson 57 pictured the mask as 1s and 0s laid over the scores. But a score of 0 isn't "no attention": e⁰ = 1, so a masked word would still get weight. For "fire" looking at "the fire was", multiplying by the mask leaves "was" a weight of 0.14, peeking at a word that hasn't been written yet. Masked scores must become −∞ before the softmax; then e^(−∞) = 0 and the weight is exactly 0.

Worked example

Worked example: −∞ against multiplying by the mask

"fire" attends to "the fire was" and must not see "was". Compare the right mask with the slip.

import numpy as np

def softmax(s):
    e = np.exp(s - s.max())
    return e / e.sum()

q = np.array([1.0, 2.0])
K = np.array([[1.0, 0.0], [0.0, 1.0], [1.0, 1.0]])
s = K @ q / np.sqrt(2)
m = np.array([1.0, 1.0, 0.0])
print(softmax(np.where(m == 1, s, -np.inf)).round(3).tolist())
print(softmax(s * m).round(3).tolist())
In the lesson you can change this and run it.

With −∞, the hidden word "was" gets exactly 0 and the visible words share everything: 0.330 and 0.670. With the slip, "was" still gets 0.140, and the visible words' weights are wrong too, since they now share only 0.86. A language model trained with the slip would learn to read the next word from the future, then fail the moment it has to write one.

Quick check
What does this print?
import numpy as np

S = np.array([[1.0, 5.0], [2.0, 2.0]])
mask = np.tril(np.ones((2, 2)))
S = np.where(mask == 1, S, -np.inf)
E = np.exp(S - S.max(axis=1, keepdims=True))
print((E / E.sum(axis=1, keepdims=True)).tolist())
  1. [[1.0, 0.0], [1.0, 0.0]]
  2. [[0.5, 0.5], [0.5, 0.5]]
  3. [[1.0, 0.0], [0.5, 0.5]]
  4. [[0.018, 0.982], [0.5, 0.5]]
Show the answer

[[1.0, 0.0], [0.5, 0.5]]

The first word may only see itself, so it takes all its weight despite the 5; the second word sees both equal scores and splits them.

Drill

Write attention

Write attention(Q, K, V, causal) that returns softmax(QKᵀ/√d)V, setting scores above the diagonal to −∞ when causal is True. Then set out_row to the masked output for word 1 ("fire") of the three-word example, rounded to 3.

import numpy as np

def attention(Q, K, V, causal):
    S = Q @ K.T
    E = np.exp(S)
    return E @ V

Q = np.array([[2.0, 0.0], [1.0, 2.0], [1.0, 1.0]])
K = np.array([[1.0, 0.0], [0.0, 1.0], [1.0, 1.0]])
V = np.array([[1.0, 0.0], [0.0, 1.0], [0.5, 0.5]])
out_row = attention(Q, K, V, True)[1].round(3).tolist()
print(out_row)
In the lesson, your code is checked as you write it.
In your own words
The first word's output is always its own value vector, whatever the scores. Why?

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

In AI and ML: what attention buys a language model

Purpose. See a word's vector absorb the words it needs.

Before attention, "warm" in "the fire was warm" is the same vector in every sentence. After attention, its new vector is a blend weighted toward the words that match it, so it carries context: in this sentence, something about a fire. In the five-word example, "warm" gives itself 0.427, then "fire" 0.259, the earlier word it shares a feature with, and 0.157 each to "the" and "was". Its new vector moves toward "fire".

The weight table, the mask laid over it. Here is all of A at once, from the Python part's own scores, S = XXᵀ/√4: one row per word, each row its weights in percent, gray where the mask hides a later word. Row "warm" reads 16, 26, 16 and 43: the 0.157, 0.259, 0.157 and 0.427 the cell printed, as percentages.

Two refinements turn this into a transformer. First, the example used X itself as queries, keys and values, so each word matched itself best (a vector's dot product with itself is its squared length). Real models multiply X by three learned matrices first, Q = XW_Q, K = XW_K and V = XW_V, so a word can look for something other than itself, a verb looking for its subject, say. Second, they run several of these "heads" side by side, each free to look for something different, and join their outputs. Each is still this lesson's eight lines. Lesson 111's training loop sets all those W's.

Here is the first refinement on the camp log. Count the features from 0, as Python does: "warm" has features 1 and 3, and "fire" has 1 and 2. This W_Q keeps features 0, 1 and 2 as they are and moves feature 3 onto feature 2, so "warm" now asks for features 1 and 2: exactly what "fire" offers. W_K and W_V are np.eye(4), the identity of lesson 57, which changes nothing, so only the queries move.

import numpy as np

X = np.array([[1.0, 0.0, 0.0, 0.0], [0.0, 1.0, 1.0, 0.0], [0.0, 0.0, 1.0, 0.0],
              [0.0, 1.0, 0.0, 1.0], [1.0, 0.0, 0.0, 1.0]])
W_Q = np.array([[1.0, 0.0, 0.0, 0.0], [0.0, 1.0, 0.0, 0.0],
                [0.0, 0.0, 1.0, 0.0], [0.0, 0.0, 1.0, 0.0]])
Q, K, V = X @ W_Q, X @ np.eye(4), X @ np.eye(4)
S = Q @ K.T / np.sqrt(4)
S = np.where(np.tril(np.ones((5, 5))) == 1, S, -np.inf)
E = np.exp(S - S.max(axis=1, keepdims=True))
A = E / E.sum(axis=1, keepdims=True)
print(Q[3].tolist(), A[3].round(3).tolist())
print((A @ V)[3].round(3).tolist())

"warm"'s query is now (0, 1, 1, 0), "fire"'s own vector, and its weights become 0.143, 0.387, 0.235, 0.235 and 0: "fire" gets the biggest share, more than "warm" itself. Its new vector, (0.143, 0.622, 0.622, 0.235), leans further toward "fire" than the 0.259 share could take it.

Quick check
Why do real models compute Q = XW_Q and K = XW_K instead of using X directly?
  1. So a word can look for something other than itself
  2. To avoid the softmax
  3. Because X can't be multiplied by itself
  4. To make the vectors shorter and cheaper to store
Show the answer

So a word can look for something other than itself

With raw X every word matches itself best. Learned projections let a query ask for, say, "my subject" and a key answer it, as "warm" just asked for "fire".

In PyTorch. The eight lines with tensors, then the projections as layers. torch.triu(…, diagonal=1) keeps the part of a matrix above its diagonal, so hide is True exactly where the mask has a 0, and masked_fill(hide, float("-inf")) writes −∞ there. F.softmax(S, dim=-1) runs across each row. Then three nn.Linear(4, 4, bias=False) layers hold W_Q, W_K and W_V. A layer stores its weights as (out, in) and multiplies by them turned around (lesson 55), so W_Q goes in as W_Q.T. Last, .double() turns the float32 results into float64, so they round and print as NumPy's do.

import torch
import torch.nn as nn
import torch.nn.functional as F

X = torch.tensor([[1.0, 0.0, 0.0, 0.0], [0.0, 1.0, 1.0, 0.0], [0.0, 0.0, 1.0, 0.0],
                  [0.0, 1.0, 0.0, 1.0], [1.0, 0.0, 0.0, 1.0]])
hide = torch.triu(torch.ones(5, 5), diagonal=1) == 1
S = (X @ X.T / 4 ** 0.5).masked_fill(hide, float("-inf"))
A = F.softmax(S, dim=-1)
print(S.shape, A.shape, (A @ X).shape)
print(A[3].double().round(decimals=3).tolist())

W_Q = torch.tensor([[1.0, 0.0, 0.0, 0.0], [0.0, 1.0, 0.0, 0.0],
                    [0.0, 0.0, 1.0, 0.0], [0.0, 0.0, 1.0, 0.0]])
to_q, to_k, to_v = nn.Linear(4, 4, bias=False), nn.Linear(4, 4, bias=False), nn.Linear(4, 4, bias=False)
with torch.no_grad():
    to_q.weight.copy_(W_Q.T)
    to_k.weight.copy_(torch.eye(4))
    to_v.weight.copy_(torch.eye(4))
Q, K, V = to_q(X), to_k(X), to_v(X)
A = F.softmax((Q @ K.T / 4 ** 0.5).masked_fill(hide, float("-inf")), dim=-1)
print(Q[3].tolist(), A[3].double().round(decimals=3).tolist())

It prints the shapes the Python part printed, 5 × 5, 5 × 5 and 5 × 4, and "warm"'s row, 0.157, 0.259, 0.157, 0.427 and 0. Then, through the three layers, the query (0, 1, 1, 0) and the weights 0.143, 0.387, 0.235, 0.235 and 0, exactly as the NumPy cell above. In a real model the three layers are wider, start random, and training sets them.

Quick check
What does this print? (Two words, queries and keys both X.)
import numpy as np

X = np.array([[1.0, 0.0], [1.0, 1.0]])
S = X @ X.T
print(S.tolist(), int(np.argmax(S[1])))
  1. [[1.0, 1.0], [1.0, 2.0]] 1
  2. [[2.0, 1.0], [1.0, 1.0]] 0
  3. [[1.0, 1.0], [1.0, 2.0]] 0
  4. [[1.0, 0.0], [1.0, 1.0]] 1
Show the answer

[[1.0, 1.0], [1.0, 2.0]] 1

Word 1's best match is itself: its dot product with itself is its squared length, 2. That's why real models learn separate W_Q and W_K.

Drill

Who does "warm" attend to?

Run masked self-attention on the five camp-log words (queries, keys and values all X). Set row to word 3 ("warm")'s weights rounded to 3, top_other to the position of the earlier word it attends to most, not counting itself, and moved to its output vector rounded to 3.

import numpy as np

X = np.array([[1.0, 0.0, 0.0, 0.0], [0.0, 1.0, 1.0, 0.0], [0.0, 0.0, 1.0, 0.0],
              [0.0, 1.0, 0.0, 1.0], [1.0, 0.0, 0.0, 1.0]])
row = []
top_other = -1
moved = []
print(row, top_other, moved)
In the lesson, your code is checked as you write it.
In your own words
"warm" attends to "fire" because they share a feature in these invented vectors. In a trained model, where would that shared feature come from?

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

Goes deeper in

Practice what you just read

  • 4 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

In attention, what does the dot product of a query and a key measure?

  1. How well the key matches what the query seeks
  2. The length of the whole sentence so far
  3. The output vector
  4. The loss

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