Series

Statistics for AI Research

A practical learning path from probability and statistical inference to ML/NLP evaluation. Each note has a separate Chinese version and a language switch.

Start with the first note
中文
Statistics posts

Part I

Statistical Foundations

  1. 1. Why Statistics Matters for AI Research
  2. 2. Random Variables, Expectation, and Variance
  3. 3. Probability Distributions for Machine Learning

Part II

Uncertainty & Model Evaluation

  1. 1. Confidence Intervals: Why One Score Is Not Enough
  2. 2. Bootstrap Resampling for Model Evaluation
  3. 3. Hypothesis Testing Without Rituals
  4. 4. Effect Size: Statistical Significance Is Not Enough

Part III

Statistics → Machine Learning

  1. 1. Linear Regression as a Statistical Model
  2. 2. Regularization: Statistics Behind L1 and L2
  3. 3. Bias–Variance Trade-off
  4. 4. PCA: From Variance to Representation
  5. 5. Maximum Likelihood Estimation: Why Models Learn Parameters
  6. 6. MAP Estimation and Bayesian Thinking

Part IV

Statistical Thinking for AI/NLP

  1. 1. Multiple Comparisons in AI Experiments
  2. 2. Power Analysis: How Many Experiments Do You Need?
  3. 3. ANOVA for Comparing Multiple Models
  4. 4. Statistical Testing for NLP and LLM Evaluation
  5. 5. Correlation vs. Causation in AI Research

How this series is designed

The series borrows the learning strengths of Seeing Theory, StatQuest, Think Stats, ISLR/ISLP, and statistical inference texts, then translates them into AI/NLP research habits.

Concept first

Each note starts with a visual mental model before formulas or experimental details.

AI/NLP examples

Examples connect the concept to MT, RAG, LLM evaluation, bias analysis, and applied ML systems.

Research writing

The notes emphasize how to make careful claims, not just how to compute a number.

Start with uncertainty.

My working definition: statistics is the language I use when an AI experiment produces a number and I need to decide how much to trust it.

Read the first note
中文版本