A neural network library written in Rust

I've been working slowly on getting solid BLAS/LAPACK abstractions.

Have you already looked into Julia? It has a lot of these things (if not all of them; I'm not sure about the ndarray bit).

I think BLAS/LAPACK are getting old, I know of cases where they got their lunch eaten by JIT'd VMs.

Much of stats and linear alg is embarrassingly vectorizable. I would love to see modern SIMD implementations (preferably written in RUST) :)

Would rust be much faster than Julia for these kinds of things?

And are compile languages even suited for notebook style programming?

In sum, everything you already have in Python.

Everything except execution times that are remotely acceptable.