Neural networks have been around since the last century but in the last decade, they have reshaped how we see the world. From classifying images of animals to extracting parts of speech, researchers are building deep neural networks in diverse and vast fields to push and break boundaries.

But as advancements in deep learning reach new heights, a new concept has lately been introduced that is a twist on the old neural network architecture – Capsule Networks. It improves on the effectiveness of the old traditional methods and understands even when presented with a situation that is shown from a different angle.

In this article, we will see why Capsule Networks are suddenly getting so much attention in the deep learning community, what the intuition behind this concept is and we will then do a code walk through to strengthen and solidify our concepts. Note: I suggest you go through the below articles before reading further in case you need to brush up your neural network concepts: You must have already heard the tales of CapsuleNets – but let us see it through with our eyes.

To compare the power of this new architecture – let us monitor CapsNet on Analytics Vidhya’s Identify the Digits Problem. Note – The code for this section is included in the “Code Walkthrough section”.

Alternatively, you can check out the code on GitHub. For the uninitiated, the Identify the Digits problem, as the name suggests, is simply a digit recognition problem. Read more from…

thumbnail courtesy of