juice

juice copied to clipboard

An attempt to resolve #12 Difficulty appears around defining a `trait BackandBounded: FeatX + FeatY` where `trait LayerOps` is backend bound. Everything else would need to depend on the `LayerOps`...

drahnr

Declare `LayerOps` per backend/framework

1

drahnr

Get rid of feature macros

1

Max: > Rust "impl specialization" should land Rust 1.9(?) (PR here: rust-lang/rust#30652). > > That feature should hopefully allow us to better determine the capabilities of the different frameworks that...

drahnr

Move `juice-examples` out from `juice` repo

7

Because cargo does not supports workspaces inside workspace, but I do want to add some more example networks (besides MNIST)

andreytkachenko

Implement maxout layer

2

See [Goodfellow 2013](http://proceedings.mlr.press/v28/goodfellow13.pdf) for paper.

subversive-owl

* [ ] lrn * [x] linear * [ ] conv

drahnr

Works for parity, should work here too https://crates.io/crates/cargo-unleash

drahnr

Native feature parity items

6

I'm curious what the current list of items needed to catch native up to the current feature set is, across all the repositories. (My ulterior motive here is not having...

subversive-owl

Currently a `gemm` operation is used to calculate the bias, which is a `O(n^3)` operation, where it should be a simply sum per `SharedTensor` element

drahnr

make layer API rustic

5

The API includes too much boilerplate, consider builder pattern.

drahnr