Hey Starlord here
With a deep learning framework, you create the training algorithm, you provide the data, then you use your data and algorithm to create your own custom model that you can host yourself.
However, this means that you need to code everything yourself as well. Fortunately for us, we now have really good free open source deep learning frameworks that allow anyone to create your own deep learning algorithms.
For example, Google’s TensorFlow, which supports Python, R, and C; and is currently the most popular of these frameworks. Microsoft cognitive toolkit, which is highly optimized and provides support for Python, C, and C#.
[KERAS logo. Keras]
Keras, which sits on top of TensorFlow to make it much easier to use.
Torch which is a Lua based framework used by Facebook, Twitter, and Google. MXNet by Apache, which provides support for R, Python, C, and Julia; and Caffe and its successor Caffe2, which provides support for both Python and C.
[TORCH logo. torch]
Now let’s take a look at a quick demo where we’ll create a deep neural network using the Keras and TensorFlow framework. So, what are the pros and cons of using a deep learning framework? On the plus side, these services are custom.
You can train your model with your own data and tune it exactly how you like it. They’re local; you can host your model on-premises, in the cloud, or embed your model right into your application, and they’re private.
If you have sensitive customer data or you’re creating proprietary technology, neither ever has to leave your data centre. On the downside, this option is complex. Building algorithms, preparing data, and training deep learning models requires a lot of knowledge and skill.
They’re labour intensive, it takes a lot of time to create the algorithms, gather and prepare the data, and train the model; and they’re expensive. The amount of time and effort required makes this by far the costliest of these three options.
However, sometimes the problem you’re attempting to solve can only be solved by this option. Ultimately, you should use this option when neither of the first two option are possible.
So, next chapter will be our final chapter of this AI journey so stay tuned.
Refrences for further reading
