\theta

Theta

Naming convention: meaning depends on the field

In plain English

A name, not an operation. By convention $\theta$ means an angle in geometry and physics, and "all of the model’s parameters" in statistics and machine learning, where it can stand for billions of numbers bundled into one symbol. Read $p_\theta(x)$ as "the probability the model with parameters $\theta$ assigns to $x$ ".

As code

// θ in ML: one name for every trainable number
const theta = { W1: [[0.1, -0.3], [0.7, 0.2]], b1: [0, 0.1], W2: [[0.5], [-0.4]] };

Where the analogy bends: θ is a variable name, like calling a variable `params`. There is no θ function.

See it

θ90° = 1.57 rad

\theta

as an angle (geometry, physics) or as the name for all of a model’s parameters (ML). Same letter, unrelated meanings, and context decides which.

What it means, field by field

Mathematics

An angle (in radians or degrees); polar coordinates (r, θ).

Physics

Angles of incidence, pendulum displacement, scattering angle.

AI / ML

The full parameter vector of a model; θ* often denotes optimal parameters.

Biology

In population genetics θ = 4Nₑμ, a scaled mutation rate. A field-specific convention again.

In a research paper

The standard statement of supervised learning (e.g. Goodfellow, Bengio & Courville, Deep Learning, 2016).

\theta^{*} = \arg\min_{\theta}\ \mathbb{E}_{(x,y)}\big[\mathcal{L}(f_\theta(x), y)\big]

The best parameters are the ones that make the average loss over the data as small as possible.

Common misreadings

Expecting θ to mean the same thing in a physics and an ML paper. It almost never does.
Imagining θ is a single number in ML. It is usually a giant vector.

Connected symbols

\arg\min

Argmin (the input that minimises)

\nabla

Gradient operator (nabla, del)

\mathcal{L}

Loss function (calligraphic L)

\eta

Eta

Where it appears on this site

∂ Gradient descent update § 2. The Physics of Systems § 3. Biology as an Information System § 6. The Mathematics Inside Modern AI § 7. The Physics–Biology–AI Connection Lab