En-NNabla-blement of answer to Google’s TensorFlow?
The libraries support static and dynamic computation graphs as well as functions, operators and optimizer modules for neural networks.
On the backend, the code is written primarily in C++11. If you want to run it on embedded devices, you can access it natively via C++11, otherwise you can use a good old, trusty Python API.
So far, Sony has tested the libraries on Ubuntu 16.04 and Windows 8 and 10, according to a blog post.
José Miguel Hernández-Lobato, a computer scientist at the University of Cambridge in the United Kingdom who specializes in machine learning, commented that Sony’s framework “seems something similar to existing frameworks” for machine learning such as TensorFlow, (Py)Torch and Theano.
“The authors do not describe in detail the differences with respect to existing tools or include comparisons to them,” he said, “so it is hard to know how useful it can be in practice.”
Some of Sony’s products that use neural networks include real estate price estimates, gesture sensitivity and image recognition. According to the blog post, the libraries are designed with “high speed and memory efficiency”.
Here’s a link if you’d like to have a play. It’s installable via
pip and there are a few Jupyter notebook tutorials available with the source code. ®