The above post was at the end of 2019 when I was using Version 12.0 . I looked at the changes in the Wolfram Neural Net Repository as Version 13.2 for Japanese was released in January 2023.
The number of models has been increased about 1.5 times 140 from 95.
In[]:=
Length@NetModel[]
Out[]=
140
The number of tasktypes has been increased from 10 to 11. Added is “Denoising”.
In[]:=
NetModel["FastDVDNet Trained on DAVIS Data","TaskType"]
Out[]=
{Denoising}
The number of layer functions used in the models has been reduced from 36 to 34. ConstantArrayLayer, ConstantPlusLayer, and ConstantTimesLayer are no longer used, and instead the new NetArrayLayer is used. ConstantcLayer and ConstantTimesLayer has been phased out in favor of NetGraph. ConstantArrayLayer has been phased out in favor of NetArrayLayer introduced at Version 12.2.
In[]:=
WolframLanguageData["NetArrayLayer","VersionIntroduced"]
Out[]=
12.2
The current Circular Diagram of the Wolfram Neural Net Repository is as follows
In Version 13.2 a new construct NetExternalObject is introduced, which makes it easy to import models from other platforms. As a result of this, I look forward to seeing more attractive ideas using Neural Network Models in the Wolfram Language.