Differentiable computing for R4TF

Skip to first unread message

Chris Black

Dec 7, 2020, 6:54:03 PM12/7/20
to Rust for TensorFlow

Are there plans to build differentiable computing into the rust for tensorflow project similiar to Swift for Tensorflow, i.e a custom derivatives for rust functions.



Adam Crume

Dec 7, 2020, 7:13:56 PM12/7/20
to Chris Black, Rust for TensorFlow
We don't currently have plans to add differentiable programming to TensorFlow Rust.  Swift's differentiable programming relies on language-level support which doesn't exist in Rust.  Graphs built with normal ops (e.g. ops::add) will have derivatives defined, although I know that's not as ergonomic as defining a function which takes e.g. f32 parameters and returns an f32.

You received this message because you are subscribed to the Google Groups "Rust for TensorFlow" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rust+uns...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/rust/e52c4d74-3824-4afb-a9a4-c380407764d0n%40tensorflow.org.
Reply all
Reply to author
0 new messages