Subsampling in Ensemble Kalman Inversion

Abstract We consider the Ensemble Kalman Inversion which has been recently introduced as an efficient, gradient-free optimization method to estimate unknown parameters in inverse setting. In case of large data sets, becomes computationally infeasible misfit needs be evaluated for each particle iteration. Here, randomised algorithms like stochastic gradient descent have demonstrated successfully overcome this issue by using only a random subset iteration, so-called subsampling techniques. Based on recent analysis continuous-time representation methods, we propose, analyse, and apply subsampling-techniques within Inversion. Indeed, propose two different techniques: either every observes same (single subsampling) or (batch subsampling).