Thermodynamic lower bounds are approximated for exact and SGD linear regression, producing energy-aware scaling laws for optimal training dataset size given a target generalization error.
The Physical Basis of the Gibbs-von Neumann entropy
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
We develop the argument that the Gibbs-von Neumann entropy is the appropriate statistical mechanical generalisation of the thermodynamic entropy, for macroscopic and microscopic systems, whether in thermal equilibrium or not, as a consequence of Hamiltonian dynamics. The mathematical treatment utilises well known results [Gib02, Tol38, Weh78, Par89], but most importantly, incorporates a variety of arguments on the phenomenological properties of thermal states [Szi25, TQ63, HK65, GB91] and of statistical distributions[HG76, PW78, Len78]. This enables the identification of the canonical distribution as the unique representation of thermal states without approximation or presupposing the existence of an entropy function. The Gibbs-von Neumann entropy is then derived, from arguments based solely on the addition of probabilities to Hamiltonian dynamics.
fields
cond-mat.stat-mech 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
The Thermodynamic Costs of Simple Linear Regression
Thermodynamic lower bounds are approximated for exact and SGD linear regression, producing energy-aware scaling laws for optimal training dataset size given a target generalization error.