Batch Normalization - Part 3: Backpropagation & Inference

[@marcoxs36]

This is the most complete explanation of BN I've seen on youtube. Thank you!

[@MLForNerds]

Glad it was helpful!

[@ThineshKumar-gk8hb]

Good explanation, efficient knowledge transfer is done, expecting more from your side like this.

[@ankurdas1477]

The video series of Batch Normalization is really gem. Before watching the series I thought that I knew everything about BN😅. Another request from my side, please complete the object detection playlist.

[@MLForNerds]

Thank you 😊

[@anantmohan3158]

Thank you for creating such wonderful videos.
The maths behind the BN was beautifully explained. Eagerly waiting for the next videos of Python and Pytorch implementation of BN.
Thank you..!

[@MLForNerds]

Thank you 👍

[@lokeshborawar9899]

@MLForNerds Why do running_mean and running_var continue to be updated during training, even after the dataset has been processed for an epoch?

[@sridharchandrasekar7787]

Sir I have one doubt, here batch refers to what...eg, in CNN , we have the Matrix of 255 X 255 ..how we calculate mean and variance

[@MLForNerds]

Batch is an additional dimension on top of 255x255 in your case. Suppose you have a grayscale image of 255x255. If there are 10 images in the batch, the dimension becomes 10x255x255. You calculate the mean and variance along batch dimension for each pixel.

[@sridharchandrasekar7787]

@@MLForNerds Batch means mini batch size?..if i give mini batch size =32...then how it will calculate sir

[@muthukamalan.m6316]

please cover group norm and layer norm as well

[@MLForNerds]

I will cover them in the coming videos. Thank you.

[@sandeepanand3834]

please add
1. forward backward prop from dropouts, maxpool, residual....
2. forward backward prop, when we use adam instead of SGD..
please