So far it seems like Google and Microsoft (and their partners - in machine learning terms - Amazon, Facebook ...) are against shipping into a machine learning web standard even basic training capabilities. But what is machine learning without training capabilities? - someone might ask. :-) Yes - it would be web machine learning standard without "learning" word in it ;-) But seriously, it would just make standard dependent from 3rd party software like ONNX or TensorFlow (of which formats are being now pushed and forced to become a standard - yes, there is a little fight between Google and Microsoft in this area :-) - wonder why? ;-) )
And if these formats will become a standard for web then there is no chance that true machine learning (with training capabilities) will be available for native web any time soon (not mentioning that ONNX and TF standards have nearly NOTHING in common with native web)
Not trying to guess why it happens, but seeing so many investments into these corporations farms of servers and products for machine learning it is hard not to have impression that this is just a part of their business strategy. Of course companies exists for profit, so nothing unusual here, but after all there wouldn't be anything surprising if ...
... providing training capabilities wasn't just about 70 lines of code. Yes, 70 lines! Of code that is ... available for free and in worst case scenario would have to be just ported from one language to another one. And not even mentioning that web standards are for everyone, not only for big corporations with particular interests.
If you read it till here and you want standard to be simple to use and to learn and have training capabilities, give a like to this post and/or share your comment (for example: "I want training capabilities" :-) ) before it is too late. Feel free to share that story with others.
Check out also these websites of web standardization group:
And these websites too (machine learning libraries made for web from scratch in Javascript):
