Symmetry in probability and statistics

Symmetry is powerful concept not just in geometry and physics but also in probability and statistics. Important manifestations of symmetry include exchangeability and invariant estimation .

Exchangeability

A random sequence is exchangeable if it is invariant in distribution under finite permutations. Pedagogical treatments of exchangeability include:

  • Aldous, 1985: Exchangeability and related topics (doi, pdf)
  • Schervish, 1995: Theory of Statistics (doi)
    • Sec. 1.2: Exchangeability & Sec. 1.4: De Finetti’s representation theorem

Symmetry and invariance

Exchangeability (permutation invariance) is not the only kind of invariance important in probability and statistics. Major references on group-theoretic and symmetry methods in probability are:

  • Diaconis, 1988: Group representations in probability and statistics (online )
  • Kallenberg, 2005: Probabilistic symmetries and invariance principles (doi)
    • Book review by Diaconis (doi, pdf)

And in statistics:

  • Eaton, 1983: Multivariate statistics: A vector space approach (online )
  • Eaton, 1989: Group invariance applications in statistics (online )
  • Kondor, 2008, PhD thesis: Group theoretical methods in machine learning (pdf)

Several MathOverflow questions turn up other interesting references (1 ,2 ).