US AI Academy

The Math Behind AI

How to read the math

The symbols in this course, said in plain words. Come back whenever one stops you.

Math notation is shorthand, nothing more. Each symbol below stands for a short sentence. You’ll meet each one again later, most with a worked example, so a quick look now is plenty.

SymbolSay itWhat it means, with an example
a1, a2, aia_1,\ a_2,\ a_i“a one”, “a sub i”The first, the second, or the i-th number in the list aa. The little letter i stands for any position. If a=(2,1,3)a=(2,1,3), then a1=2a_1=2 and a3=3a_3=3.
w1x, 2bw_1x,\ 2b“w one x”, “two b”Two things written side by side are multiplied. So w1xw_1x means w1×xw_1\times x. And 2b2b means 2×b2\times b.
a⋅b\mathbf a\cdot\mathbf b“a dot b”The dot product of the lists a\mathbf a and b\mathbf b: multiply them pair by pair, then add (lesson 3).
WijW_{ij}“W i j”The number in row i, column j of the grid WW.
∑\sum“sum” (capital sigma)Add them all up. Small print under and over it says where to start and stop: ∑i=13ai=a1+a2+a3=2+1+3=6\sum_{i=1}^{3}a_i=a_1+a_2+a_3=2+1+3=6.
ee“e”A fixed number, about 2.718, the way π\pi is about 3.142. Your calculator’s exe^x key raises ee to a power: e2≈7.389e^2\approx7.389.
23, ex2^3,\ e^{x}“2 to the 3rd”, “e to the x”Multiply that many copies of the number together: 23=2×2×2=82^3=2\times2\times2=8. Anything to the power 0 is 1. A minus sign in the power means “1 divided by”: 2−3=1/82^{-3}=1/8. For powers that aren’t whole numbers, like e1.5e^{1.5}, use your calculator’s key.
log⁡2, ln⁡\log_2,\ \ln“log base 2”, “L N”Undo a power. log⁡28\log_2 8 asks “2 to what power gives 8?” The answer is 3. ln asks the same about ee. Most calculators have ln but no log₂ key, so use log⁡2x=ln⁡x÷ln⁡2\log_2 x=\ln x\div\ln 2.
 , <, >\sqrt{\ },\ \lt,\ \gt“square root of”, “less than”, “more than”16=4\sqrt{16}=4, because 4×4=164\times4=16. 2<32\lt3 and 3>23\gt2; a negative number is <0\lt0.
∣a∣|\mathbf a|“the length of a”How long the arrow a\mathbf a is: square each number, add them up, take the square root. ∣(3,4)∣=9+16=5|(3,4)|=\sqrt{9+16}=5.
θ, φ\theta,\ \varphi“theta”, “phi”Greek letters used as names for angles.
cos⁡, cos⁡−1\cos,\ \cos^{-1}“cosine”, “inverse cosine”cos turns an angle into a number from −1 to 1. cos⁻¹ turns it back; it’s also called arccos or acos, and here the −1 is part of the name, not “1 divided by”. For example, cos⁡60∘=0.5\cos 60^\circ=0.5, cos⁡−10.5=60∘\cos^{-1}0.5=60^\circ. Set the calculator to DEG for degrees or RAD for radians. One radian is about 57.3°.
0.5×(2,4)0.5\times(2,4)a number times a listMultiply each number in the list: 0.5×(2,4)=(1,2)0.5\times(2,4)=(1,2). Two lists add position by position: (1,2)+(3,0)=(4,2)(1,2)+(3,0)=(4,2).
K⊤K^{\top}“K transpose”Write each row of the grid KK as a column: row 1 becomes column 1, row 2 becomes column 2, and so on.
σ(z)\sigma(z)“sigma of z” (small sigma)The sigmoid, a smooth yes or no. It turns any number into a chance between 0 and 1, and σ(0)=0.5\sigma(0)=0.5. The formula is σ(z)=1/(1+e−z)\sigma(z)=1/(1+e^{-z}).
y^, t\hat y,\ t“y-hat”, “t”y^\hat y is the network’s answer; tt is the truth it should have said.
←\leftarrow“becomes”Replace the left side with the right side. w←w−1w\leftarrow w-1 means w goes down by 1.
η\eta“eta”The step size used in training.
∂A∂B\dfrac{\partial A}{\partial B}“the slope of A with respect to B” (experts say “the partial”)How fast AA changes when BB moves a tiny bit, with everything else held still. You can check one yourself: nudge BB by 0.001, see how much AA moves, and divide by 0.001. You’ll get very nearly the slope.
δ\delta“delta”One neuron’s shared slope, used in backpropagation (idea 5).
⇒, ≈\Rightarrow,\ \approx“so”, “about”⇒\Rightarrow: so it follows that. ≈\approx: approximately equal.

The two sigmas are different letters: big ∑\sum adds things up, small σ\sigma is the sigmoid.