-
Automatic differentiation of truncated SVD
Truncated SVD (Singular Value Decomposition) is an important linear algebra operation that has wide applications in various tensor network algorithms. In this article, we will study the (reverse mode) automatic differentiation of truncated SVD. Du...…
-
On the back-propagation of CG linear system solvers
We have seen from the previous article that the dominant eigen-decomposition as a function primitive for AD needs CG or other linear system solver for its backward pass. In this article, we will furthermore investigate the feasibility of carrying ...…
-
The adjoint method and its application in back-propagation of dominant eigen-decomposition
The adjoint method provides a new way of thinking about and deriving formulas for certain computation process. In this article, we will present the basic ideas of the adjoint method and, as a concrete example, demonstrate its consequences in back...…