Index Ai Academy
Foundation · Putting it together

Residual connections and layer norm

Adding the input back and why its slope keeps learning alive, layer norm by hand, and add and norm.

About 40 minutes · Free lesson

What it covers6

Why it matters

Attention (lesson 113) and a small network (lesson 110) are two pieces of a transformer block. Two more pieces glue them together, and both are short. The first adds each piece's input back onto its output: x + f(x). It looks like nothing, but its slope has a 1 in it, and that 1 is why networks dozens of layers deep can learn at all. The second rescales each word's vector to mean 0 and deviation 1, the standardizing of lesson 97 done one vector at a time, so the numbers stay a steady size from layer to layer. Together they're called "add and norm".

What you'll be able to do

  • Compute x + f(x) and its slope, 1 + f′(x), and say why the added 1 keeps slopes alive through many layers
  • Standardize one vector to mean 0 and deviation 1, then apply a learned scale γ and shift β
  • Tell layer norm from batch norm by which direction is averaged, and catch the slip of averaging over the batch
  • Follow the shapes through add and norm, and get the same numbers from PyTorch's nn.LayerNorm
Quick check
Have a guess before you start. Start with 1 and halve it, twenty times in a row. About how much is left?
  1. About 0.000001
  2. About 0.1
  3. Exactly 0
  4. 10
Show the answer

About 0.000001

0.5²⁰ is about 0.00000095: halve anything twenty times and almost nothing is left. That's what happens to a slope passing back through twenty layers that each halve it, and this lesson's first idea is the fix.

The math: add the input back, then rescale

Purpose. Two short steps that let a network stack many layers: add each layer's input back, and keep every vector the same size.

Adding the input back. Any layer, a small network or attention, takes a vector x and gives back f(x). A residual connection sends x around the layer and adds it back on:

out=x+f(x)ddx(x+f(x))=1+f′(x)\text{out} = x + f(x) \qquad \frac{d}{dx}\big(x + f(x)\big) = 1 + f'(x)

In plain words: keep what came in, and add the layer's change to it. The slope is the slope of x, which is 1, plus the slope of f: the sum rule of lesson 73.

The layer now only has to learn a change, and the 1 in the slope turns out to matter most.

Quick check
From lesson 74: a path runs through three steps with slopes 2, 0.5 and 3. What is the slope of the whole path?
  1. 1
  2. 6
  3. 5.5
  4. 3
Show the answer

3

Along a chain, slopes multiply: 2 × 0.5 × 3 = 3. That's why many layers in a row can shrink a slope to almost nothing: every layer multiplies it again.

Why the 1 matters. A slope passing back through twenty layers is the product of their twenty slopes. A plain chain needs every slope close to 1 to survive, and that's hard to keep: twenty slopes of 0.5 leave 0.5²⁰, about 0.00000095, so the first layers barely learn. In a residual chain each factor is 1 + f′ instead, so each layer's own part only needs a slope close to 0. That's easy: a layer whose weights start small changes its input only a little. Twenty factors of 1.02 make about 1.49, a usable size. The 1 is not magic, though: twenty factors of 1.5 make about 3,325. Residual blocks work because each one's own part stays small.

The picture. Here is one residual layer, drawn as a calculation. Its own part halves its input, f(x) = 0.5x, and x = 2 goes two ways: through f, and around it. Press Next to send the slope back.

Try it now

The thick route around the layer carries the slope back multiplied by 1, whatever f does.

Rescaling each vector: layer norm. After an add, a vector's numbers can be any size, and they drift from layer to layer. Layer norm standardizes each vector on its own, as lesson 97 standardized a column: subtract its mean μ, divide by its deviation σ. Then it scales each entry by a learned γ (gamma) and shifts it by a learned β (beta), so training can give each entry back whatever size and center works best.

The picture. Step through add and norm for one word's vector, x = (2, 1, 1, 0), after a sublayer handed back f(x) = (2, 0, −1, 1). The bars are the vector's four entries, drawn on one scale for the whole flow, so you can watch them move: the add, the mean, the centering, the shrink to deviation 1, and last γ and β, which this layer has learned as (2, 1, 1, 1) and (0, 0, 0.5, 0).

Symbols and formula. For a vector y with d entries:

μ=1d∑i=1dyiσ=1d∑i=1d(yi−μ)2+εouti=γi yi−μσ+βi\mu = \frac{1}{d}\sum_{i=1}^{d} y_i \qquad \sigma = \sqrt{\frac{1}{d}\sum_{i=1}^{d} (y_i - \mu)^2 + \varepsilon} \qquad \text{out}_i = \gamma_i \, \frac{y_i - \mu}{\sigma} + \beta_i

In plain words: center the vector on 0, shrink or stretch it to deviation 1, then let each entry take its own learned scale and shift.

ε (epsilon) is a tiny number, 0.00001 in PyTorch, so a vector whose entries are all equal, with σ = 0, never divides by 0. The flow above leaves it out to keep the hand arithmetic exact. γ and β have one number per entry, so a layer norm for vectors of 4 learns 8 numbers. A new one starts at γ = 1 and β = 0: pure standardizing, until training moves them.

Quick check
A vector is (5, 5, 5, 9). What is its mean, and what does its first entry become after subtracting it?
  1. 5 and 0
  2. 6 and 1
  3. 6 and −1
  4. 24 and −19
Show the answer

6 and −1

(5 + 5 + 5 + 9) ÷ 4 = 24 ÷ 4 = 6, and 5 − 6 = −1. The four centered entries, −1, −1, −1 and 3, add to 0.

In your own words
In your own words: why does adding x back make a deep network easier to train?

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

In Python: slopes through twenty layers, and layer norm in three lines

Purpose. Measure the slope through twenty layers both ways, then write layer norm and check it.

def slope_through(f_slope, layers, residual):
    total = 1.0
    for layer in range(layers):
        if residual:
            total = total * (1 + f_slope)
        else:
            total = total * f_slope
    return total

print(f"{slope_through(0.5, 20, False):.1e}", f"{slope_through(0.5, 20, True):.0f}")
print(f"{slope_through(0.02, 20, False):.1e}", f"{slope_through(0.02, 20, True):.2f}")

Each layer multiplies the slope again, by f′ in a plain chain or by 1 + f′ with a residual. With f′ = 0.5, the plain chain falls to 9.5e-07 and the residual one blows up to 3325. With f′ = 0.02, a layer that changes little, the plain chain is gone, 1.0e-34, while the residual one stays a usable 1.49. Residuals turn "keep every slope near 1", which is hard, into "keep every change small", which small starting weights already do.

Now layer norm, three lines on the flow's x + f(x). np.allclose asks whether two numbers agree to about five places, as math.isclose did in lesson 7:

import numpy as np

y = np.array([4.0, 1.0, 0.0, 1.0])
mu = y.mean()
sigma = np.sqrt(((y - mu) ** 2).mean() + 1e-5)
y_hat = (y - mu) / sigma
print(mu, round(sigma, 4), y_hat.round(4).tolist())
print(np.allclose(y_hat.mean(), 0), np.allclose(y_hat.std(), 1))

The mean is 1.5, σ rounds to 1.5 (the 1e-5 under the root is ε), and ŷ is (1.6667, −0.3333, −1.0, −0.3333), the flow's numbers. True and True: mean 0 and deviation 1, up to float rounding.

Code flow. A model norms a whole batch of vectors at once, one per row. Step through three rows. axis=1 averages across each row, and keepdims=True keeps one number per row as a column, so it lines up with the row it came from. The second row is the first times ten.

A named mistake: averaging over the batch. axis=0 averages each column down the batch instead of across each vector. That's a real method, batch norm, used in networks for pictures, but it isn't layer norm: each word's numbers would then depend on whatever else is in the batch.

import numpy as np

Y = np.array([[4.0, 1.0, 0.0, 1.0], [40.0, 10.0, 0.0, 10.0], [2.0, 2.0, 3.0, 1.0]])
right = (Y - Y.mean(axis=1, keepdims=True)) / Y.std(axis=1, keepdims=True)
slip = (Y - Y.mean(axis=0)) / Y.std(axis=0)
print(right[0].round(4).tolist())
print(slip[0].round(4).tolist())

Across the row, the first vector comes out as before. Down the columns, it comes out as something else, set by the other two rows: change them and it changes too. A transformer reads one word at a time when it writes, with no batch to average over, so it norms each vector on its own.

Worked example

Worked example: the whole of add and norm

Write layer norm with its learned γ and β as a function, and run the flow's add and norm through it.

import numpy as np

def layer_norm(v, gamma, beta):
    mu = v.mean()
    sigma = np.sqrt(((v - mu) ** 2).mean() + 1e-5)
    return gamma * (v - mu) / sigma + beta

x = np.array([2.0, 1.0, 1.0, 0.0])
fx = np.array([2.0, 0.0, -1.0, 1.0])
gamma = np.array([2.0, 1.0, 1.0, 1.0])
beta = np.array([0.0, 0.0, 0.5, 0.0])
out = layer_norm(x + fx, gamma, beta)
print(out.round(4).tolist(), out.shape)
In the lesson you can change this and run it.

It prints (3.3333, −0.3333, −0.5, −0.3333), the flow's answer, and the shape (4,): a vector of 4 went in and a vector of 4 comes out. γ doubled the first entry and β lifted the third by 0.5. The order matters: add first, then norm, so the residual's sum is what gets rescaled.

Quick check
What does this print?
import numpy as np

v = np.array([3.0, 3.0, 5.0, 5.0])
print(v.mean(), v.std(), ((v - v.mean()) / v.std()).tolist())
  1. 4.0 1.0 [3.0, 3.0, 5.0, 5.0]
  2. 4.0 1.0 [-1.0, -1.0, 1.0, 1.0]
  3. 4.0 2.0 [-0.5, -0.5, 0.5, 0.5]
  4. 16.0 1.0 [-1.0, -1.0, 1.0, 1.0]
Show the answer

4.0 1.0 [-1.0, -1.0, 1.0, 1.0]

The mean is 4, every entry is 1 away from it, so σ = 1, and standardizing gives −1, −1, 1, 1.

Drill

Layer norm for a batch

Fix layer_norm(Y) so it standardizes each row of Y on its own: mean 0 and deviation 1 across the row, with 1e-5 under the square root. The starter has three slips.

import numpy as np

def layer_norm(Y):
    mu = Y.mean(axis=0)
    var = ((Y - mu) ** 2).mean(axis=0)
    return (Y - mu) / var

Y = np.array([[4.0, 1.0, 0.0, 1.0], [2.0, 2.0, 3.0, 1.0]])
out = layer_norm(Y)
print(out.round(4).tolist())
In the lesson, your code is checked as you write it.
In your own words
Why does layer norm average across each vector rather than down the batch?

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

In AI and ML: add and norm in a transformer block

Purpose. Follow the shapes through add and norm, and run both halves in PyTorch.

Shapes through add and norm. A transformer block does add and norm twice: once after attention, once after the small network. Take the five camp-log words of lesson 113, "the fire was warm tonight", each 4 numbers wide: X is 5 × 4. Attention gives back one vector per word, 5 × 4; the add needs the two shapes equal, and gives 5 × 4; and layer norm rescales each row, each word, on its own: 5 × 4 again. Add and norm never change the shape, which is what lets a block stack on a block.

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]])
S = X @ X.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))
H = X + (E / E.sum(axis=1, keepdims=True)) @ X
mu = H.mean(axis=1, keepdims=True)
out = (H - mu) / np.sqrt(((H - mu) ** 2).mean(axis=1, keepdims=True) + 1e-5)
print(X.shape, H.shape, out.shape)
print(H[3].round(3).tolist(), out[3].round(4).tolist())
print(np.allclose(out.mean(axis=1), 0), out.std(axis=1).round(4).tolist())

"warm" was (0, 1, 0, 1); attention added (0.157, 0.686, 0.416, 0.427), so H's row is (0.157, 1.686, 0.416, 1.427). Layer norm turns that into (−1.1796, 1.1796, −0.78, 0.78). Every row now has mean 0 and deviation 1.0.

Quick check
X is 5 × 4. Suppose a sublayer gave back 5 × 16. Can a residual add the two?
  1. Yes: NumPy stretches the smaller one
  2. No: x + f(x) needs both the same shape
  3. No: a residual only works after attention
  4. Yes: the add makes it 5 × 20
Show the answer

No: x + f(x) needs both the same shape

The residual adds entry by entry, so f(x) must come back the same shape as x. That's why a transformer's small network widens to 16 and then narrows back to 4 before its add.

In PyTorch. nn.LayerNorm(4) is layer norm for vectors of 4: its weight is γ, its bias is β, and it adds ε = 1e-5 under the root. A residual connection is just a plus sign: x + layer(x). Here the layer is nn.Linear(1, 1, bias=False) with its one weight set to 0.5, so it halves its input, like the picture's f.

import torch
import torch.nn as nn

norm = nn.LayerNorm(4)
y = torch.tensor([2.0, 1.0, 1.0, 0.0]) + torch.tensor([2.0, 0.0, -1.0, 1.0])
print(norm(y).double().round(decimals=4).tolist())
with torch.no_grad():
    norm.weight.copy_(torch.tensor([2.0, 1.0, 1.0, 1.0]))
    norm.bias.copy_(torch.tensor([0.0, 0.0, 0.5, 0.0]))
print(norm(y).double().round(decimals=4).tolist())

layer = nn.Linear(1, 1, bias=False)
with torch.no_grad():
    layer.weight.fill_(0.5)
x = torch.tensor([2.0], requires_grad=True)
out = x + layer(x)
out.backward()
print(layer(x).item(), out.item(), x.grad.item())

A new nn.LayerNorm starts at γ = 1 and β = 0, so its first answer is the three NumPy lines' (1.6667, −0.3333, −1.0, −0.3333). With the flow's γ and β loaded, it gives the flow's (3.3333, −0.3333, −0.5, −0.3333). Then autograd (lesson 82) finds the residual's slope for us: f(x) = 1.0, x + f(x) = 3.0, and x.grad = 1.5, which is 1 + f′. .double() turns float32 into float64 so the numbers round and print as NumPy's do.

The gradient reaching the first of twenty layers. Now real layers: twenty nn.Linear(4, 4) layers whose weights start small, made with NumPy's random generator so every run gets the same ones. The same layers are stacked twice, plain and with residuals, and backward() sends a slope from the top to the input x:

import numpy as np
import torch
import torch.nn as nn

rng = np.random.default_rng(0)
layers = []
for k in range(20):
    layer = nn.Linear(4, 4, bias=False)
    with torch.no_grad():
        layer.weight.copy_(torch.tensor(rng.normal(0, 0.1, (4, 4))))
    layers.append(layer)

for residual in [False, True]:
    x = torch.ones(4, requires_grad=True)
    h = x
    for layer in layers:
        if residual:
            h = h + layer(h)
        else:
            h = layer(h)
    h.sum().backward()
    print(residual, f"{x.grad.norm().item():.1e}")

x.grad.norm() is the size of the slope that reached x (lesson 49). Through the plain stack, the False line, it's 2.5e-15: nothing to learn from. Through the same layers with residuals, the True line, it's 1.3e+00, which is 1.3. One plus sign per layer is the whole difference.

Quick check
What does this print?
import torch

x = torch.tensor(3.0, requires_grad=True)
out = x + 0.25 * x
out.backward()
print(out.item(), x.grad.item())
  1. 3.75 0.25
  2. 0.75 1.25
  3. 3.75 1.25
  4. 3.75 1.0
Show the answer

3.75 1.25

out = 3 + 0.75 = 3.75, and its slope is 1 + 0.25 = 1.25: the 1 from x, the 0.25 from f(x) = 0.25x.

Drill

Add and norm for a sentence

Write add_norm(X, F) that adds the sublayer's output F to X and then layer-norms each row (1e-5 under the root). Then set shapes to the shapes of X, F and the result, and centered to whether every row of the result has mean 0 (use np.allclose).

import numpy as np

def add_norm(X, F):
    Y = (X - X.mean(axis=1, keepdims=True)) / X.std(axis=1, keepdims=True)
    return Y + F

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]])
F = np.array([[0.5, 0.5, 0.0, 0.0], [0.2, 0.1, 0.4, 0.3], [0.0, 0.3, 0.3, 0.0]])
out = add_norm(X, F)
shapes = []
centered = False
print(out.round(4).tolist(), shapes, centered)
In the lesson, your code is checked as you write it.
In your own words
A transformer block is attention, add and norm, a small network, add and norm. Why can it be stacked dozens of times?

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

Goes deeper in

Practice what you just read

  • 7 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 residual layer computes x + f(x). What is its slope?

  1. 1 + f′(x)
  2. f′(x)
  3. x × f′(x)
  4. 1

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