AbstractIn this short article we suggest randomized scalable stochastic matrix-based algorithms for large linear systems. The idea behind these stochastic methods is a randomized vector representation of matrix iterations. In addition, to minimize the variance, it is suggested to use stochastic and double stochastic matrices for efficient randomized calculation of matrix iterations and a random gradient based search strategy. The iterations are performed by sampling random rows and columns only, thus avoiding not only matrix matrix but also matrix vector multiplications. Further improvements of the methods can be obtained through projections by a random gaussian matrix.
