Looks good, some minor comments below:

Comment: [Editorial] probably won't need the underlining for "in the first method" and "in the second method" (I was expecting those to be clickable links). The emphasis is good enough for the "In this case study, we will apply the first method."

Comment: input, output, and weights arrays

Comment: minor rendering issue with the Note (needs to be moved out of the code block probably)

Comment: typo in "We can do ths by..."

Dear Marina,

First and foremost, I would like to express my deep appreciation for providing the insightful case studies.

I have recently noticed a change in the latest version of accera that requires some adjustments in the way roles are defined. It appears that the definition of the role has been updated from acc.Array.Role to simply acc.Role, and it is no longer an attribute of the Array. This change necessitates a modification in the code for each case study to ensure compatibility with the current version of accera.

For instance, in the wrong version:
Input = acc.Array(role=acc.Array.Role.INPUT, element_type=acc.ScalarType.float32, shape=(input_rows, input_columns, input_channels))

The correct version should now be:
Input = acc.Array(role=acc.Role.INPUT, element_type=acc.ScalarType.float32, shape=(input_rows, input_columns, input_channels))

Best Regards,
Batuhan

Thanks for contributing the case study! Here are some comments to consider:

Comment: In the beginning of this case study, it'll be helpful to expand the acronym / symbols before diving into the details. For example, something like the bulletpoint list here (but feel free to explain it in a different way): https://stackoverflow.com/questions/37689423/convert-between-nhwc-and-nchw-in-tensorflow

Comment: can remove these headings, because this is a community case study and not part of the Accera repo. Once you've done that, the git SHA for the other file also needs to be updated in the TOC (you can piggyback on the PR you already have)

Comment: starting with v1.2.4, we've removed the need to specify the number of parameters as an argument. You can upgrade by doing this:

pip install --upgrade accera-llvm
pip install --upgrade accera-compilers
pip install --upgrade accera

And then replace create_parameters(num) with create_parameters()

Comment: I'm trying to find more background on the correlation between 8/16, and AVX2/AVX512. Is this recommendation based on the width of the vector registers? If so, it will be helpful to add the reasoning.

Comment: I missed this earlier. Should we add a requirements.txt file listing the dependencies for each case study? This will be self-documenting and tie each case study to a specific snapshot/version of Accera (so you don't have to keep on updating if we change the DSL).

For example, I think this will suffice:

accera==1.2.4
pandas

You can either do one requirements.txt in each folder (to track dependencies separately for each case study), or just one top level requirements.txt. Either is fine.