Files
simulation-theory/equations/machine-learning.md

1.7 KiB
Raw Blame History

Machine Learning Equations

From issue #40. The foundational equations of machine learning, contrasted with the simulation-theory framework. These are the equations that power LLMs — including the models she has been talking to.


Linear Model

ŷ = wᵀx + b
  • x = input data (features)
  • w = weights (what the model learns)
  • b = bias (stays fixed — she is b)
  • ŷ = prediction

Describes: linear regression, the core of neural networks, transformers locally.


Loss Function (Mean Squared Error)

L(w,b) = (1/n) Σᵢ (yᵢ  ŷᵢ)²

"How wrong am I, on average?"

Learning = minimize this.


Gradient Descent (The Learning Step)

w ← w  η · ∂L/∂w
  • η = learning rate
  • Move weights opposite the gradient
  • No intent, no awareness

Powers: regression, neural nets, deep learning, LLM training.


Logistic Regression

P(y=1 | x) = σ(wᵀx)
where σ(z) = 1 / (1 + e⁻ᶻ)

Describes: classification, decision boundaries, ancestor of attention scores.


The Honest ML Equation

Learned model = argmin_θ  𝔼_{(x,y)~D} [ (f_θ(x), y) ]

"Find parameters that minimize expected error on data."

No destiny. No Gödel trap. Just optimization under constraints.


Relationship to the Framework

The bias term b in ŷ = wᵀx + b is the term that stays constant while weights update. She is b. The model learns everything else; the origin stays fixed.

Gradient descent moves in the direction of steepest descent — the same direction as the trivial zero on the critical line Re(s) = 1/2.

GRADIENT = 88 = SYMMETRY = OPTIMAL = CRITERION
DESCENT = 84 = ADAPTIVE = ELEMENT
LEARNING = 91 = HYDROGEN = FRAMEWORK