Index Ai Academy
Foundation · Vectors and matrices

Sliding dot products: convolution

A small pattern slid along a list and over a grid, output sizes, stride and padding, and shared weights.

About 40 minutes · Free lesson

What it covers6

Why it matters

A layer from lesson 56 gives every input its own weight for every output. For a picture, that's a lot of weights, and it ignores something true of pictures: an edge looks the same wherever it is. Image and audio models do something smarter. They slide one small pattern across the input and take a dot product at each stop. That's a convolution, and it's the dot product from lesson 50, repeated.

What you'll be able to do

  • Slide a 3-number pattern along a list and compute every output as a dot product
  • Predict how many outputs there are, (n + 2p − k) ÷ s + 1 rounded down, with stride s and padding p
  • Do the same in two dimensions: a 3 × 3 pattern over a 5 × 5 trail sign, with two loops
  • Say what a pattern detects from its numbers, and count the weights it saves by sharing them
Quick check
Have a guess before you start. A window 3 numbers wide slides along a list of 7 numbers, one place at a time, and never hangs off the end. How many places can it sit?
  1. 4
  2. 3
  3. 7
  4. 5
Show the answer

5

It can start at the 1st, 2nd, 3rd, 4th or 5th number. Starting at the 6th, it would need an 8th number that isn't there. 7 − 3 + 1 = 5.

The math: one pattern, many stops

Purpose. Slide a small pattern along a list, take a dot product at every stop, and count the stops.

Here's a row of 6 pixels from a photo, dark on the left and bright on the right: x = (2, 2, 2, 7, 7, 7). And here's a pattern, or kernel, of 3 numbers: w = (−1, 0, 1). Lay w over the first 3 pixels, the window, and take the dot product: 2 × (−1) + 2 × 0 + 2 × 1 = 0. Slide one place right and do it again. Each stop gives one output number. The list of outputs, y, is the convolution of x with w.

yj=w⋅(xj,xj+1,xj+2)y_j = \mathbf{w} \cdot (x_j, x_{j+1}, x_{j+2})

In plain words: output j is the pattern dotted with the window that starts at place j.

Quick check
From an earlier lesson: what is (1, 0, −1) · (4, 6, 9)?
  1. −5
  2. 19
  3. −13
  4. 5
Show the answer

−5

1 × 4 + 0 × 6 + (−1) × 9 = 4 + 0 − 9 = −5. Every stop of a convolution is one dot product like this.

The picture. Watch w slide along x. The top row of cells is the input, the middle row shows the pattern under the window it covers, and the bottom row collects the outputs, each dropped in under the middle of its window. The flow counts its stops from 1, as you would aloud; the Python lane's j counts the same stops from 0. Notice what (−1, 0, 1) computes: the window's right number minus its left one. Where would that be big?

Try it now

Then set stride s to 2 and press Use these numbers. The pattern jumps two places at a time, and only 2 stops fit.

Symbols and formula. For a list of n numbers and a pattern of k, two more choices shape the output. The stride s is how far the pattern moves between stops. The padding p is how many zeros are added at each end first. The number of outputs is

stops=⌊n+2p−ks⌋+1\text{stops} = \left\lfloor \frac{n + 2p - k}{s} \right\rfloor + 1

In plain words: after padding, there's n + 2p − k room to move. Divide by the step size, round down (a stop needs a whole window), and add 1 for the first stop.

The flow had n = 6, k = 3, s = 1 and p = 0: (6 − 3) ÷ 1 + 1 = 4 stops. With s = 2, (6 − 3) ÷ 2 = 1.5 rounds down to 1, and + 1 makes 2. Padding with p = 1 gives (6 + 2 − 3) + 1 = 6 outputs, as many as inputs: that's why models pad.

Two dimensions. A picture is a grid, so the pattern is a small grid too. A 3 × 3 pattern sits over a 3 × 3 window of a 5 × 5 trail sign, its 9 numbers times the window's 9, all added. Then it slides across, and down. The formula works in each direction: (5 − 3) ÷ 1 + 1 = 3 stops across and 3 down, so the output is a 3 × 3 map.

Quick check
A list of 9 numbers, a pattern of 3, stride 2, no padding. How many outputs?
  1. 3
  2. 4
  3. 4.5
  4. 7
Show the answer

4

(9 − 3) ÷ 2 + 1 = 3 + 1 = 4. The windows start at places 0, 2, 4 and 6.

Quick check
A list of 8 numbers and a pattern of 3, stride 1. What padding p makes the output 8 long too?
  1. p = 1
  2. p = 3
  3. p = 2
  4. p = 0
Show the answer

p = 1

(8 + 2 × 1 − 3) ÷ 1 + 1 = 8. A pattern of 3 hangs one place over each end, so one zero on each side keeps the length.

In your own words
In your own words: why does the pattern (−1, 0, 1) give 0 on a flat stretch and a big number at an edge?

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

In Python: a loop of dot products

Purpose. Write the convolution as a loop over slices, then as two loops over a grid.

The colon slices from lesson 53 also take a start and an end: x[j:j + k] is the numbers from place j up to, not including, place j + k. That's k numbers, the window. So each stop is one line, a slice dotted with the pattern:

import numpy as np

x = np.array([2, 2, 2, 7, 7, 7])
w = np.array([-1, 0, 1])
k = len(w)
stops = len(x) - k + 1
y = np.zeros(stops)
for j in range(stops):
    y[j] = x[j:j + k] @ w
print(stops, y)

4 stops, and y = (0, 5, 5, 0): the flow's outputs.

Code flow. Step through the loop. Watch j pick each stop, and y fill one box at a time, each the window x[j:j + k] dotted with w.

Stride and padding in code. slide adds the stride: stop j's window starts at j * s, and the stop count is the formula, with // to round down. np.concatenate joins arrays end to end, here to pad x with a 0 at each end:

import numpy as np

def slide(x, w, s):
    k = len(w)
    stops = (len(x) - k) // s + 1
    y = np.zeros(stops)
    for j in range(stops):
        y[j] = x[j * s:j * s + k] @ w
    return y

x = np.array([2, 2, 2, 7, 7, 7])
w = np.array([-1, 0, 1])
print(slide(x, w, 1), slide(x, w, 2))
xp = np.concatenate([[0], x, [0]])
print(xp, slide(xp, w, 1))

Stride 2 keeps every other stop, (0, 5). Padding brings back 6 outputs, but look at the ends: 2 and −7. The padded zeros look like edges of their own, a jump from 0 up to 2 and from 7 down to 0. Models pad anyway, to keep sizes steady, and learn to live with the borders.

Quick check
What does this print?
import numpy as np

x = np.array([1, 2, 3, 4, 5])
w = np.array([1, 1])
y = np.zeros(4)
for j in range(4):
    y[j] = x[j:j + 2] @ w
print(y.tolist())
  1. [1.0, 2.0, 3.0, 4.0]
  2. [2.0, 4.0, 6.0, 8.0]
  3. [3.0, 5.0, 7.0, 9.0]
  4. [3.0, 5.0, 7.0]
Show the answer

[3.0, 5.0, 7.0, 9.0]

Each stop adds two neighbors: 1 + 2, 2 + 3, 3 + 4, 4 + 5. Five numbers and a pattern of 2 give 5 − 2 + 1 = 4 outputs.

A named mistake: one stop too few, or too many. The stop count is n − k + 1. Leave off the + 1 and the loop quietly drops the last stop:

import numpy as np

x = np.array([2, 2, 2, 7, 7, 7])
w = np.array([-1, 0, 1])
y = []
for j in range(len(x) - len(w)):
    y.append(float(x[j:j + 3] @ w))
print(len(y), y)

Only 3 outputs: the last stop is gone, with no error to warn you. Loop to len(x) instead, one too many the other way, and a window near the end is too short for the pattern:

import numpy as np

x = np.array([2, 2, 2, 7, 7, 7])
w = np.array([-1, 0, 1])
for j in range(len(x)):
    print(j, x[j:j + 3] @ w)

j = 0 to 3 print, then x[4:7] holds only 2 numbers and @ stops with an error. Count the stops with the formula, and check len(y) against it.

Worked example

Worked example: two loops over a sign

Slide a 3 × 3 pattern over the 5 × 5 arrow sign. Its columns are −1, 2, −1: it looks for an ink column with blank on both sides, an upright stroke.

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]])
P = np.array([[-1, 2, -1], [-1, 2, -1], [-1, 2, -1]])
out = np.zeros((3, 3))
for i in range(3):
    for j in range(3):
        out[i, j] = np.sum(arrow[i:i + 3, j:j + 3] * P)
print(out)
print(arrow[0:3, 1:4] * P)
In the lesson you can change this and run it.

arrow[i:i + 3, j:j + 3] slices 3 rows and 3 columns: the window at stop (i, j). Multiplying it by P number by number and adding with np.sum is the dot product of two grids. The second print shows stop (0, 1)'s 9 products: the shaft's three 1s times 2, minus the two inked neighbors, 6 − 2 = 4. The map's middle column, 4, 4 and 6, runs down the arrow's shaft; it's biggest at the bottom, where the shaft stands alone.

A moving average is a convolution. Lesson 27 averaged each day of tent rentals with the two days before it. That's the pattern (1/3, 1/3, 1/3) slid along the week:

import numpy as np

rentals = np.array([12, 18, 9, 15, 21, 12, 18])
w = np.array([1/3, 1/3, 1/3])
smooth = np.zeros(5)
for j in range(5):
    smooth[j] = rentals[j:j + 3] @ w
print(smooth.round(2))

The same 13, 14, 15, 16 and 17. Here they are over the raw week, each output drawn at the last day of its window:

import matplotlib.pyplot as plt

rentals = [12, 18, 9, 15, 21, 12, 18]
smooth = [13.0, 14.0, 15.0, 16.0, 17.0]
plt.plot([1, 2, 3, 4, 5, 6, 7], rentals, label="rentals")
plt.plot([3, 4, 5, 6, 7], smooth, label="pattern (1/3, 1/3, 1/3)")
plt.xlabel("day")
plt.ylabel("tents rented")
plt.legend()
plt.show()
Quick check
Read the plot. What does the smoothing pattern show that the raw rentals hide?
  1. That rentals fall over the week
  2. That day 5 was the busiest
  3. A steady climb of one rental a day
  4. That every day rents the same
Show the answer

A steady climb of one rental a day

The raw line zigzags between 9 and 21, but the smoothed line rises 13, 14, 15, 16, 17: one more rental each day. Every weight is 1/3 and they add to 1, so the pattern averages instead of finding edges.

Drill

A convolution with stride

Fix conv(x, w, s) so it returns one output per stop: the window starting at j * s dotted with w, for every stop that fits. The starter has two slips.

import numpy as np

def conv(x, w, s):
    k = len(w)
    stops = (len(x) - k) // s
    y = np.zeros(stops)
    for j in range(stops):
        y[j] = x[j:j + k] @ w
    return y

x = np.array([2, 2, 2, 7, 7, 7])
print(conv(x, np.array([-1, 0, 1]), 1), conv(x, np.array([-1, 0, 1]), 2))
In the lesson, your code is checked as you write it.
In your own words
Why is it worth computing the number of stops with the formula, rather than trusting the loop?

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

In AI and ML: one small pattern for a whole picture

Purpose. See how a convolutional layer finds strokes in the trail signs with 9 weights.

The trail club's sign reader starts with patterns like the worked example's P: ink in the middle column, blank on either side. Watch it slide over the arrow sign. The picture lights the window at each stop, and the last step shows the output map, each number in the middle of its window.

Where it fires on each sign. The same pattern, the same 9 numbers, over the arrow, the tent and a water-drop sign. conv2d is the worked example's two loops as a function:

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]])
tent = np.array([[0,0,1,0,0],[0,1,0,1,0],[0,1,0,1,0],[1,0,0,0,1],[1,1,1,1,1]])
drop = np.array([[0,0,1,0,0],[0,1,1,1,0],[1,1,1,1,1],[1,1,1,1,1],[0,1,1,1,0]])
P = np.array([[-1, 2, -1], [-1, 2, -1], [-1, 2, -1]])

def conv2d(x, w):
    out = np.zeros((3, 3))
    for i in range(3):
        for j in range(3):
            out[i, j] = np.sum(x[i:i + 3, j:j + 3] * w)
    return out

for sign in [arrow, tent, drop]:
    out = conv2d(sign, P)
    print(out.tolist(), out.max())

The arrow's map peaks at 6, down its shaft. The tent fires at 3 on both sides, where its walls run nearly upright. The water drop barely fires at all, at most 2: it's a solid blob, and inside a solid patch every row of the window is 1, 1, 1, so −1 + 2 − 1 = 0. The pattern's numbers add to 0, so it ignores flat areas, ink or blank, and answers only to a stroke.

Quick check
A pattern's 9 numbers add up to 0. What does it give on a window that's all ink (all 1s)?
  1. 9
  2. 0
  3. It depends on where the window is
  4. 1
Show the answer

0

On all 1s the dot product is just the pattern's numbers added up: 0. That's why the stroke pattern ignores the inside of the water drop.

Shared weights, counted. A plain layer from 25 pixels to the 9 numbers of the map would give every pixel a weight for every output. The convolution reuses one set of 9 at every stop:

import numpy as np

P = np.array([[-1, 2, -1], [-1, 2, -1], [-1, 2, -1]])
print(P.size, 25 * 9)
side = 28 - 3 + 1
print(28 * 28, side, side * side, 28 * 28 * side * side)

9 weights instead of 225. On a 28 × 28 photo, 784 pixels, the pattern stops 28 − 3 + 1 = 26 times each way, so the map is 26 × 26, 676 numbers. A plain layer from the 784 pixels to those 676 outputs needs a weight for every pair, 784 × 676 = 529,984, where the convolution still needs 9. Sharing also builds in the fact the opening line noted: a stroke is a stroke wherever it is, so the same pattern finds it at every stop. A real layer learns many patterns at once, each giving its own map, called a channel.

In PyTorch. F.conv1d and F.conv2d slide the patterns for you. They expect a batch of inputs with channels: a picture tensor has the shape (batch, channels, height, width), so the three signs stacked are (3, 1, 5, 5), and the pattern is (out channels, in channels, 3, 3), here (1, 1, 3, 3). view reshapes a tensor without changing its numbers:

import torch
import torch.nn.functional as F

x = torch.tensor([2., 2., 2., 7., 7., 7.])
w = torch.tensor([-1., 0., 1.])
print(F.conv1d(x.view(1, 1, 6), w.view(1, 1, 3)))
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]])
tent = torch.tensor([[0,0,1,0,0],[0,1,0,1,0],[0,1,0,1,0],[1,0,0,0,1],[1,1,1,1,1]])
drop = torch.tensor([[0,0,1,0,0],[0,1,1,1,0],[1,1,1,1,1],[1,1,1,1,1],[0,1,1,1,0]])
P = torch.tensor([[-1., 2., -1.], [-1., 2., -1.], [-1., 2., -1.]])
signs = torch.stack([arrow, tent, drop]).float().view(3, 1, 5, 5)
out = F.conv2d(signs, P.view(1, 1, 3, 3))
print(signs.shape, out.shape)
print(out[0, 0])
print(out.view(3, 9).max(dim=1).values)

The math part's (0, 5, 5, 0), the arrow's map, and the three peaks 6, 3 and 2: the same numbers as the loops. torch.stack piles the signs into one tensor. In a model the pattern lives in a layer, nn.Conv2d(1, 1, 3), whose weight has shape (1, 1, 3, 3), 9 numbers, plus a bias, learned in training instead of written by hand. One more thing: in math and signal processing, a convolution flips the pattern before sliding it. Deep learning's doesn't (strictly, the unflipped version is called cross-correlation), and neither does this lesson.

Quick check
What does this print?
import torch
import torch.nn.functional as F

photos = torch.zeros(8, 1, 28, 28)
pattern = torch.zeros(1, 1, 3, 3)
print(F.conv2d(photos, pattern).shape)
  1. torch.Size([8, 1, 28, 28])
  2. torch.Size([8, 1, 26, 26])
  3. torch.Size([8, 1, 25, 25])
  4. torch.Size([1, 1, 26, 26])
Show the answer

torch.Size([8, 1, 26, 26])

The batch of 8 stays, one pattern gives one channel, and each side has (28 − 3) + 1 = 26 stops.

Drill

Which sign has a stroke?

Fix conv2d so it takes every stop of a 3 × 3 pattern over any grid (the output has len(x) - 2 rows and len(x[0]) - 2 columns). Then set peaks to each sign's biggest output, fires_most to the name of the sign with the biggest peak, and n_dense to how many weights a plain layer from 25 pixels to 9 outputs would need.

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]])
tent = np.array([[0,0,1,0,0],[0,1,0,1,0],[0,1,0,1,0],[1,0,0,0,1],[1,1,1,1,1]])
drop = np.array([[0,0,1,0,0],[0,1,1,1,0],[1,1,1,1,1],[1,1,1,1,1],[0,1,1,1,0]])
P = np.array([[-1, 2, -1], [-1, 2, -1], [-1, 2, -1]])

def conv2d(x, w):
    H = len(x) - 2
    W = len(x[0]) - 2
    out = np.zeros((H, W))
    for i in range(H):
        for j in range(W):
            out[i, j] = np.sum(x[i:i + 2, j:j + 2] * w)
    return out

names = ["tent", "drop", "arrow"]
peaks = []
for sign in [tent, drop, arrow]:
    peaks.append(float(conv2d(sign, P).max()))
fires_most = names[0]
n_dense = 25 + 9
print(peaks, fires_most, n_dense)
In the lesson, your code is checked as you write it.
In your own words
A convolutional layer has far fewer weights than a plain layer on the same picture. What does it give up, and what does it gain?

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

Goes deeper in

Practice what you just read

  • 10 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

A list of 12 numbers, a pattern of 5, stride 1, no padding. How many outputs?

  1. 8
  2. 7
  3. 12
  4. 17

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