Beyond IGO-Flow: Toward Convergence Analysis of IGO in Continuous Spaces
Researchers have developed a new theoretical framework for analyzing the convergence of Information-Geometric Optimization (IGO) in discrete, continuous spaces. This work focuses on IGO updates within the multivariate Gaussian family for strongly convex quadratic objectives, incorporating full covariance adaptation and a fixed learning rate. The analysis demonstrates that the covariance matrix converges to zero while the mean vector converges to the global optimum under specific conditions, advancing the theoretical understanding of IGO and its relation to practical methods like CMA-ES. AI