[rspeer]

Is a single SGD layer a neural net? Is an image filter or an audio filter a neural net? Is matrix multiplication a neural net? This would strain the intended definition even farther than it's already been strained.

But all of those are differentiable programming, and rightly so because they're all pieces that you use and compose together to make interesting learning mechanisms, including the ones that we vaguely refer to as "deep learning" now.

I like the terminology. It's not about what the original long-abandoned motivation for the design was ("neural"). It's not about how gratuitously complex you can make it ("deep"). "Differentiable" is about how it works and how we design it.

[nightski]

I'm not really buying your argument here. Neural networks are just a collection of artificial neurons. There is no requirement for multiple layers or depth of any kind.

[contrarian_]

> artificial neurons

Differentiable functions

[clarry]

Basically you're implying binary neurons, neuroevolution (which works on non-differentiable functions) etc. aren't a thing. Or at least they're not (working with) neural nets.

It's almost like SGD just made decades of AI research into neural networks just vanish.

[rspeer]

I don't see that implication at all.

Nobody is claiming that the definition of "differentiable programming" should be identical to the definition of "neural net". The claim is, if you want to assign a name to the thing that TensorFlow, PyTorch, and similar frameworks do, it's "differentiable programming".

If you want to make a non-differentiable neural net, knock yourself out. The research still exists and nobody is stopping you.

But while we're talking about terminology, I'd encourage you to stop referring to the units as "neurons". The false analogy to biology just confuses people.