Maximum likelihood estimation of log-affine models using reaction networks

Log-affine statistical models include log-linear models with prominent examples such as graphical models and hierarchical models in contingency tables. We study the inference problem of maximum likelihood estimation for log-affine models from the perspective of chemical reaction networks. For any model design matrix, we construct an appropriate mass-action system with initial concentrations given by the observed relative frequencies, such that it has the maximum likelihood estimate (MLE) as its unique positive steady state. This is joint work with Oskar Henriksson, Jose Rodriguez and Polly Yu.