核心概念
CoRMF is a novel RNN-based efficient Ising model solver that leverages criticality-ordered spin sequences to efficiently probe the generally intractable Ising model with probabilistic inference.
要約
CoRMF introduces a criticality-ordered spin sequence for N-spin Ising models, enabling unification between variational mean-field and RNN. The method is well-modularized, model-independent, and applicable to forward Ising inference problems. CoRMF optimizes the autoregressive factorization using an RNN and variance-reduced Monte Carlo gradient estimator. The framework demonstrates utility on various Ising datasets, providing tighter error bounds than naive mean-field methods.
統計
CoRMF solves the Ising problems in a self-train fashion without data/evidence.
The negative-log-partition or variational free energy of CoRMF is restricted by tighter error bound than Naive Mean-Field for general Ising graphs.
引用
"Our method has two notable characteristics: leveraging the approximated tree structure of the underlying Ising graph and being well-modularized."
"CoRMF provides an efficient surrogate model for NP problems with guaranteed local minimum convergence."
"The framework demonstrates utility on various Ising datasets, providing tighter error bounds than naive mean-field methods."