Some libraries are not doing machine learning on their own, but provide important low-level primitives for it. An example of such a library is Apple BNNS. It is a part of the Accelerate framework which provides highly optimized subroutines for convolutional neural networks. We'll discuss it in much detail in Chapter 10, Natural Language Processing. In the following section, we'll list some third-party libraries of its kind.
Low-level subroutine libraries
Eigen
Eigen is a C++ template library implementing linear algebra primitives and related algorithms. It is under the LGPL3+ license. Many popular computationally-heavy projects (TensorFlow, for instance) rely on it for matrix and vector operations.
Official...