The first-class auto-differentiation seems to me like the special bit in Julia for machine learning (not only neural networks). Flux/Zygote is only one of the t

The first-class auto-differentiation seems to me like the special bit in Julia for machine learning (not only neural networks). Flux/Zygote is only one of the tools for that (but it is indeed what I view as most promising at the moment). Do check out cassette as well. However, for the restricted set of autodiff supported by libraries like Tensorflow, basically any of the Julia tools are good enough.