We discuss the application of Riemannian optimization algorithms on smooth manifolds to the problem of approximating a given tensor by one of low (canonical) rank. The proposed algorithm incorporates local information about the condition number and can outperform state-of-the-art methods for small-scale dense tensors especially when the norms of the rank-1 tensors are different.
References:
1. https://epubs.siam.org/doi/10.1137/17M1142880
2. https://epubs.siam.org/doi/10.1137/17M114618X
Copyright. All Rights Reserved.