The notions of partition functions, entropy, free energy and variational methods prevalent in the graphical models literature come from statistical physics.

True if the history of science shows that partition functions, entropy, free energy and variational methods, as used in graphical models, came from statistical physics (Gibbs/Boltzmann distributions, mean-field and variational free energy, Bethe approximations). Mostly true if most of these came from statistical physics but one has a significant partly separate lineage, for example entropy through Shannon's information theory or variational methods through the calculus of variations, as long as the statistical-physics origin of the concepts as used in graphical models still holds. Mixed if only about half of the listed notions can be traced to statistical physics. Mostly false if most of them came mainly from other fields. False if none did.

Set before the research began.

1. E1 Advances in Variational Inference arXiv (ar5iv rendering), authors from Microsoft Research, KTH, UC Irvine

   “The origins of VI date back to the 1980s. Mean field methods, for instance, have their origins in statistical physics, where they played a prominent role in the statistical mechanics of spin glasses [121, 147].”

2. E2 An Introduction to Variational Methods for Graphical Models Machine Learning journal (Kluwer), by Jordan, Ghahramani, Jaakkola, Saul · 1999

   “In keeping with statistical mechanical terminology we will refer to this sum as a “partition function.””

3. E3 An Introduction to Variational Methods for Graphical Models Machine Learning journal (Kluwer), by Jordan, Ghahramani, Jaakkola, Saul · 1999

   “The terminology comes from the roots of the techniques in the calculus of variations.”

4. E4 Entropy in thermodynamics and information theory - Wikipedia Wikipedia · 5 December 2005

   “Because the mathematical expressions for information theory developed by Claude Shannon and Ralph Hartley in the 1940s are similar to the mathematics of statistical thermodynamics worked out by Ludwig Boltzmann and J. Willard Gibbs in the 1870s, in which the concept of entropy is central, Shannon was persuaded to employ the same term 'entropy' for his measure of uncertainty.”

E2 says the "partition function" in graphical models is named in keeping with statistical-mechanics terminology. E1 says mean-field methods have their origins in statistical physics, where they played a prominent role in the statistical mechanics of spin glasses. Two notions also have partly separate lineages. E3 says the variational terminology comes from the calculus of variations, and E4 says information-theoretic entropy was developed by Shannon, who borrowed the name from Boltzmann–Gibbs thermodynamics. No item speaks directly to free energy, but the evidence still mostly supports a statistical-physics origin, which fits the criteria for mostly true.