Amol Profile picture
1 Dec, 12 tweets, 4 min read
👋 CNN get easier to understand if you consider how we solve word puzzles (An illustrative thread) 👇🏾

#100DaysOfMLCode Image
Neural Networks came back into limelight when Prof. Yan Lecun introduced a mathematical operation, called convolution, in one of the layers of a digit recognition model. For NNs, it just turned the world upside down
Most of the AI Applications that we see today are based on convolutional neural networks (CNN).
Convolution op or use of kernels is at the center of CNNs but can be daunting to understand at first. But word puzzles offer an easy way to think about them.
In word puzzles like this, you are expected to find words. The word could are either on horizontal, vertical or cross lines. Go ahead try to find words in this one Image
The first thing I usually do is look for vowels scanning through each block Image
Once I have vowels, I look for adjacent blocks. This helps finding 2-letter words. This time we are looking 2x2 blocks at a time Image
No luck with 2-letter words. We move search to 3x3 blocks. Now are looking for neighbours' of neighbour (of vowels we found in step 1) Image
There they are. We found 2 words. Image
The 2x2 and 3x3 block perimeter - that we set for our search - is how kernel works in convolution operation.
In summary,
- Kernels are short filters to look for a specific pattern at each step
- The initial layers identify high level features while deeper layers find features more closer to the actual object Image

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Amol

Amol Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!