In this paper a sampled second-order stochastic algorithm (SSN-Lissa) is presented by combining the Lissa and SSN algorithms for large-scale machine learning optimizations. The linear convergence of the algorithm is established under the assumption that the objective functions are smooth and strongly convex. Numerical experiments show that the new algorithm is more effective than the Lissa and SSN.

Machine learning, Convex optimization, First-order stochastic method, Second-order stochastic method, Taylor approximation