EigenVI is an eigenvalue-based framework for score-based variational inference. Here we consider variational families built from orthogonal function expansions. Minimizing a stochastic estimate of the a score-based divergence (the Fisher divergence) reduces to solving an eigenvalue problem.
In this repository, we provide simple examples of EigenVI using normalized Hermite polynomials.
EigenVI is described further in the paper:
Diana Cai, Chirag Modi, Charles C. Margossian, Robert M. Gower, David M. Blei, Lawrence K. Saul. EigenVI: score-based variational inference with orthogonal function expansions. NeurIPS, 2024.