FAIRiCUBE / resource-metadata

manage information for processing/analysis resources, specifically: issue form to collect md requirements, issue template to manage codelists
https://fairicube.github.io/resource-metadata/
0 stars 0 forks source link

Architecture codelist - replace 'cnn' with 'dnn' #2

Closed jetschny closed 1 year ago

jetschny commented 1 year ago

Change proposal description

Codelist name

Architecture

Codelist value

cnn

Codelist value description

why only Convolutional neural networks?

Issue faced

too specfic

Change proposer

dnn deep neural networks

References

cozzolinoac11 commented 1 year ago

Hi @jetschny, I agree with your observation. I have already edited the value by replacing "cnn" with "dnn" to make it less specific. Thank you!

cozzolinoac11 commented 1 year ago

Hi @jetschny, the value 'cnn' was again added to the codelist of the architecture field along with other architectures to better specify the type of neural network. Therefore, now both 'dnn' and 'cnn' and other architectures are in the codelist.

KathiSchleidt commented 1 year ago

@jetschny Can this issue be closed as resolved?

jetschny commented 1 year ago

I still dont quite understand why we want to differentiate between dnn og cnn (as I beleive cnns can be seen as a subgroup of dnns). from the user persective DNN might be enough? for a datascientist CNN might add info but not a lot?!

cozzolinoac11 commented 1 year ago

In my opinion, it makes sense to provide as many details as possible when specifying the architecture used and its algorithm because this helps in understanding the resource. Now, both values ('cnn' and 'dnn') are present in the codelist of the 'Architecture' field, so we can use either the more specific or the less specific one depending on the situation.

KathiSchleidt commented 1 year ago

Maybe I'm just stupid, but if the catalog also wants to address non-ML-experts, maybe full text in place of acronyms would help? CNN I can now parse, DNN I'm still lost.

Remember that we also want domain scientists like Martin to be able to find something useful!

cozzolinoac11 commented 1 year ago

@KathiSchleidt I agree with you and we can use both the name and the acronym. For example 'CNN - Convolutional-Neural-Network' instead of just the CNN acronym. What do you think?

KathiSchleidt commented 1 year ago

Perfect! Best of both worlds!!! :)

cozzolinoac11 commented 1 year ago

I have just updated the codelist using the structure <acronym - full name>. Specifically, I updated the values of the fields 'architecture' and 'algorithm'.