2021 BRN Discussion, page-27304

that's been the basis of the metaTF environment and documentation.

a large part of the learning documentation takes very familiar test datasets for tensor flow and shows you a before and after scenario. with and without the Akida final layer. Some are greatly improved, others not so much.

the biggest improvements look to be in classification eg facial recognition, but it's less effective in the regression type modeling scenarios. that makes sense given how it works.

although I've only been playing with it, there is a clear benefit on the edge of using existing CNN's to provide the model and then releasing the model + akida on the task out in the field.

the best way I can describe this is Akida sort of turns the contrast dial up. for those old enough to remember black and white tv, sometimes the picture was a bit fuzzy and by turning the contrast dial to max, you might lose detail, but could differentiate between objects.

it's much harder to test metaTF as a pure SNN not gotten that far yet.