Just like if you were going to learn Python for Deep Learning, you're going to need 3 things:

1. Understanding of Deep Learning

2. General familiarity with the language

3. Understanding of packages/tools that will help you accomplish your goals. If this were python we'd be talking Numpy/TensorFlow/PyTorch/Keras/etc.

We're talking C++, I'm guessing you'll want to work with less abstraction than TF and the like, so BLAS, LAPACK, and maybe the whole CUDA family (CUDA, cuBLAS, cuDNN, etc).

Otherwise, TensorFlow, Caffe, CNTK, and other popular deep learning frameworks are available for C++, even though they might be better known and more heavily used in other languages.