On sparsity-exploiting memory-efficient trust-region regularized nonlinear least squares algorithms for neural-network learning | IEEE Conference Publication | IEEE Xplore