A novel machine learning method is proposed that combines tensor neural networks and a posteriori error estimators to efficiently solve high-dimensional boundary value problems and eigenvalue problems associated with second-order elliptic operators.
