Parallel implementation of the algorithm for solving entropy-robust estimation problem on heterogeneous computer systems

Paper is devoted to development of parallel technologies for numerical solving of problems to entropy-robust estimation of randomized models’ characteristics under limited amount of data. We propose to use an iterative algorithm based on batch Monte Carlo iterations. The algorithm being very computational intensive, potentially possesses high level of parallelism. Its implementation is aimed to modern computer systems with massive-parallel processing. The implementation is aimed to MPI+OpenMP+CUDA technology stack, which can be effectively mapped to modern heterogeneous computer architectures. Workability and efficiency of proposed technologies have been approved by experiments on test problem.