📝 Greek Letters
Greek symbols are conventions, not fixed definitions. Their meaning depends on the surrounding paper, code, or model, and uppercase and lowercase forms can mean different things.
The IPA below reflects common English mathematical pronunciation; regional variants are noted where especially common.
| Greek letter | English name | Pronunciation (IPA) | Common ML meanings | Example / notes |
|---|---|---|---|---|
| Α α | alpha | /ˈæl.fə/ | Learning rate, mixing coefficient, distribution shape parameter | α = 10⁻³ may be a learning rate |
| Β β | beta | US /ˈbeɪ.tə/, UK /ˈbiː.tə/ | Regression coefficient, Beta-distribution parameter, momentum coefficient | β is often a learned linear-model coefficient |
| Γ γ | gamma | /ˈɡæm.ə/ | RL discount factor, kernel scale, normalization scale | γ = 0.99 is common in reinforcement learning |
| Δ δ | delta | /ˈdel.tə/ | Change, difference, residual, Kronecker delta | Δw can denote a parameter update |
| Ε ε, ϵ | epsilon | /ˈep.sɪ.lɒn/ | Numerical-stability constant, small perturbation, RL exploration rate | ε = 10⁻⁸ can prevent division by zero |
| Ζ ζ | zeta | /ˈzeɪ.tə/ | Rare generic parameter | It has no widely fixed ML meaning |
| Η η | eta | /ˈeɪ.tə/ | Learning rate | ηₜ often denotes the learning rate at step t |
| Θ θ, ϑ | theta | /ˈθeɪ.tə/ | Model parameters | θ commonly represents all learnable parameters |
| Ι ι | iota | /aɪˈoʊ.tə/ | Rare index or indicator | It has no widely fixed ML meaning |
| Κ κ | kappa | /ˈkæp.ə/ | Cohen's kappa metric, concentration parameter | κ is the concentration parameter of a von Mises–Fisher distribution |
| Λ λ | lambda | /ˈlæm.də/ | Regularization coefficient, eigenvalue, rate parameter | L + λ‖w‖² adds an L2 penalty |
| Μ μ | mu | UK /mjuː/, US /muː/ | Mean, momentum coefficient | x ∼ 𝒩(μ, σ²) uses μ as the mean |
| Ν ν | nu | /nuː/ | Degrees of freedom, ν-SVM parameter | Its meaning is usually defined locally |
| Ξ ξ | xi | /ksaɪ/ | Random variable, noise sample, generic vector | ξ ∼ 𝒩(0, I) can denote sampled noise |
| Ο ο | omicron | /ˈɑː.mɪ.krɑːn/ | Rare generic parameter | Often avoided because it resembles Latin o and zero |
| Π π, ϖ | pi | /paɪ/ | Class prior, probability distribution, RL policy; product (uppercase) | πθ(a ∣ s) is a policy's action probability |
| Ρ ρ, ϱ | rho | /roʊ/ | Correlation coefficient, density or ratio | ρ(X, Y) can denote correlation |
| Σ σ, ς | sigma | /ˈsɪɡ.mə/ | Standard deviation, sigmoid function; summation or covariance (uppercase) | p = σ(z) can be a probability, but σ alone does not mean probability |
| Τ τ | tau | /taʊ/ | Temperature, time constant, time step | softmax(z/τ) uses temperature τ |
| Υ υ | upsilon | /ˈʌp.sɪ.lɒn/ | Rare generic variable | It has no fixed common ML meaning |
| Φ φ, ϕ | phi | /faɪ/ | Feature map, probability density or CDF, generic function | φ(x) may be a feature map; Φ(x) often denotes a normal CDF |
| Χ χ | chi | /kaɪ/ | Chi-square statistic or distribution | χ² is used in some statistical tests |
| Ψ ψ | psi | /psaɪ/ | Generic function; occasionally the digamma function | Its meaning is usually defined locally |
| Ω ω | omega | /oʊˈmeɡ.ə/ | Generic variable; sample space or parameter set (uppercase) | It has no universal ML meaning |