A Layman’s Introduction to Recurring Neural Networks.

Source: Deep Learning on Medium


The more I learn about different deep learning architectures the more I get intrigued about the different use cases for each such architecture. In terms of providing a solution for sequential data (Temporal (or) Non – Temporal) RNNs work their charm like no other.

What is Sequence Data?

Any kind of Data that has multiple steps in which there exists an order weather temporal or not, without which there cannot be a significant inference made is called a Sequence Data.

Use cases where Sequence Data is used?

  • Video — Action Recognition
  • Text — Named Entity Recognition
  • Text — Sentiment Analysis
  • Audio — Speech Recognition
  • Audio — Machine Translation

Recurring Neural Nets – Explained.

RNNs at their core are pretty simple. You just have a single weight matrix that you repeatedly use to determine the output at a step, This output will be used by the same Weight Matrix along with an external Input to evaluate the Output at the step that follows. This process repeats itself till there are no external Inputs i.e… till the end of the sequence.

Hope that made sense! Nevertheless the point of this blog is to help you understand what I understood. Hence, of all the resources that I have gone through I found this video to be very lucid and informative. It gives you a quick understanding of what exactly happens in RNNs.