# Why Gram matrixes give a good sense of style?

**URL:** <https://community.deeplearning.ai/t/why-gram-matrixes-give-a-good-sense-of-style/6717>\
**Category:** Convolutional Neural Networks\
**Tags:** coursera-platform\
**Created:** [May 17, 2021, 3:26am UTC](https://community.deeplearning.ai/t/why-gram-matrixes-give-a-good-sense-of-style/6717 "2021-05-17T03:26:57Z")\
**Posts on this page:** 1\
**Showing post:** 2

<div class="post-metadata">

**Author:** ![reinoudbosch](https://avatars.discourse-cdn.com/v4/letter/r/22d042/32.png) [@reinoudbosch](https://community.deeplearning.ai/u/reinoudbosch)\
**Post date:** [May 18, 2021, 9:45pm UTC](https://community.deeplearning.ai/t/why-gram-matrixes-give-a-good-sense-of-style/6717/2 "2021-05-18T21:45:47Z")

</div>

Hi Jaskerat,

Each layer in the channel extracts a particular feature. So to find correlations between features, values in different channels have to be compared.  
See, e.g., [this](https://community.deeplearning.ai/t/week-2-what-is-the-meaning-of-axis-3-in-the-batchnormalization/589/6) post.

As to your second question, if activations are similar, you will get a high value. If they are different they will more or less cancel each other out. In this, normalization helps to limit absolute values.

---

_[View the full topic](https://community.deeplearning.ai/t/why-gram-matrixes-give-a-good-sense-of-style/6717)._