AI glossary: terms starting with B

the backward pass on a computation graph, each local derivative used once

Example: how deep learning libraries compute gradients

start from the output and multiply local derivatives back along each edge

Example: ∂f/∂a = 10 · 1 · 3 = 30

every point within distance r of a centre

Example: ‖x − c‖₂ ≤ r

The amount of smoothing in a KDE. seaborn scales it with bw_adjust.

a mean

Example: x̄ = (2, 2), ȳ = 4.6667

Magnitude represented by length from the baseline, which is why a bar axis starts at zero.

The part of a recursive function that returns without calling the function again: if n <= 1: return 1.

Event fraction in a stated starting population: 6,636 / 30,000 = 22.12%.

in bⁿ, b is the base and n counts how many times it is multiplied

Example: 2³ = 2 × 2 × 2

A simple prediction rule used for comparison; here, the training mean for each weekday.

A transformed feature used as a column.

e₁ = (1, 0) and e₂ = (0, 1); every point is a combination of them

Example: (1, 2) = e₁ + 2e₂

Every update uses every training row.

The number of rows in a mini-batch.

The update from a prior and the likelihoods in each class to a posterior: likelihood times prior, divided by the total probability of the evidence.

Choosing each next setting from the scores of the settings tried so far.

One recorded zero or one under this model.

The probability that a transaction has fraud label one.

One modelled outcome that is one with probability p and zero otherwise.

The fewest operations any input of size n needs: a linear search that finds its value first time makes one comparison.

A probability distribution over a rate from 0 to 1, with mean a / (a + b).

Error from a model too simple for the pattern.

The two parts of the error that the penalty trades.

An asymptotic upper bound on growth. These lessons use the tightest usual bound, dropping constants and smaller terms: 4n + 4 is O(n).

A declared interval of values. [5, 10) includes 5 and excludes 10.

Classification with two classes, such as spam or ham.

Finding a value in a sorted list by comparing it with the middle of the range, and keeping the half that can still hold it.

To add a name to the namespace, with a reference to an object.

Number of ones across a fixed number of Bernoulli trials.

Probability pattern of the count across a fixed number of independent Bernoulli trials.

n times the one-trial probability p.

halve an interval that contains a zero, keeping the half where the sign changes

Example: 2x + eˣ = 0 on [−1, 0]

A view of two variables together, such as each daily count against its date.

The indented lines that belong to an if, else, while, for or case. The first line back at the outer level ends it.

True or False. The type of True and False, both keywords, written with a capital letter.

Boolean array selecting elements where it is true

Drawing new samples, with replacement, from the one observed set itself.

The collection of a statistic from many bootstrap replicates.

One resampled dataset, drawn with replacement from the observed rows.

A sample of the same size drawn from the data with replacement.

A plot of quartiles, median, whiskers and flagged observations.

One of the blocks a condition chooses between. In an if and else, exactly one branch runs.

Ends the loop at once; the program continues with the first line after it. It ends only the loop it is directly in.

The line where the debugger stops. breakpoint() in the code sets one; so does a click beside a line number in VS Code.

Matching dimensions from the right, repeating a size-one dimension conceptually

One slot in a hash table. A key's hash, modulo the number of buckets, chooses its bucket.

Contiguous storage for the array's elements

Code that runs, and does something other than what it should.