Abstract
This study investigates the dynamics of alternating minimization applied to abilinear regression task with normally distributed covariates, under theasymptotic system size limit where the number of parameters and observationsdiverge at the same rate. This is achieved by employing the replica method to amulti-temperature glassy system which unfolds the algorithm's time evolution.Our results show that the dynamics can be described effectively by atwo-dimensional discrete stochastic process, where each step depends on allprevious time steps, revealing the structure of the memory dependence in theevolution of alternating minimization. The theoretical framework developed inthis work can be applied to the analysis of various iterative algorithms,extending beyond the scope of alternating minimization.