Paul Hirschi
  • Posts
  • Portfolio
  • Projects
  • About
ABOUT

Paul Hirschi

I build models that turn noisy data into decisions — and I write about the math that makes them trustworthy.

I'm a data scientist working across applied mathematics and statistics, from Bayesian methods and experimentation to the numerical guts of the algorithms everyone reaches for. Most of my work lives where a clean derivation meets a messy dataset.

This site is where I think out loud: posts that work through the math end to end, and a portfolio of engagements where that math moved a real number. If something here is useful, wrong, or worth arguing about, get in touch.

portrait
DAILY PUZZLE · CHESS.COM
CURRENTLY READING · HARDCOVER
Cover of Optimal IllusionsOptimal IllusionsCoco Krumme2023 ↗

KEY EQUATIONS · APPLIED MATH

P(A \mid B) = \dfrac{P(B \mid A)\,P(A)}{P(B)}

Bayes’ Theorem

Inverts conditional probability to update beliefs as evidence arrives — the foundation of statistical inference.

\int_a^b f'(x)\,dx = f(b) - f(a)

Fundamental Theorem of Calculus

Ties differentiation and integration together, turning accumulated rates of change into exact totals.

f(x) = \dfrac{1}{\sigma\sqrt{2\pi}}\,\exp\!\left(-\dfrac{(x-\mu)^2}{2\sigma^2}\right)

Normal Distribution

The bell curve that sample averages converge to — the default model for noise and measurement error.

\sqrt{n}\,\big(\bar{X}_n - \mu\big) \;\xrightarrow{\,d\,}\; \mathcal{N}\!\big(0,\,\sigma^2\big)

Central Limit Theorem

Sample means look Gaussian no matter the population — the license behind most of applied inference.

\hat{\boldsymbol{\beta}} = \big(X^{\top}X\big)^{-1} X^{\top} y

Ordinary Least Squares

The closed-form solution to linear regression — the workhorse that fits a line through data.

\theta_{t+1} = \theta_t - \eta\,\nabla_{\!\theta}\,\mathcal{L}(\theta_t)

Gradient Descent

Steps downhill along the gradient to minimize loss — the optimizer behind nearly all machine learning.

\hat{\theta} = \operatorname*{arg\,max}_{\theta}\ \prod_{i=1}^{n} p(x_i \mid \theta)

Maximum Likelihood Estimation

Picks the parameters that make the observed data most probable — the dominant principle for fitting models.

H(X) = -\sum_{i} p_i \log p_i

Shannon Entropy

Measures the uncertainty in a distribution — the basis of information theory, coding, and decision trees.

f(x) = \sum_{n=0}^{\infty} \dfrac{f^{(n)}(a)}{n!}\,(x-a)^n

Taylor Series

Approximates any smooth function by polynomials — the engine of numerical methods and local analysis.

e^{i\pi} + 1 = 0

Euler’s Identity

Unites e, i, π, 1, and 0 in a single line — calculus’s gateway to Fourier analysis.

Paul Hirschi — data science · applied mathematics · statistics

 
  • github

  • scholar

  • email