-
Currently, it's hardcoded that we're working only with `numpy.float64` types. However, Firedrake can be compiled to work with complex numbers as well.
I should look into what's needed to be done to e…
-
This thread -- as a continuation of the discussion started in JuliaManifolds/Manifolds.jl#27 -- shall collect ideas and approaches to the Riemannian Gradient (for example employing tools like the `eg…
-
User worried about optimal performance will not be able to get it with a column-oriented tabular format in any event, as MultivariateStats likes observations as columns (and by definition, observation…
-
I'm playing around with NeuralODEs and adjoints right now, starting from [the official Lux.jl "MNIST Classification using Neural ODEs" tutorial](https://lux.csail.mit.edu/dev/tutorials/intermediate/1_…
-
Comparing large sparse matrices to their adjoint is very slow. It can be much slower than subtracting them.
```Julia
julia> A = sparse(Diagonal(ones(10000)));
julia> A==A';
julia> @time A==A';
…
-
If you provide both an element-wise function `f` and a dimension specification, `mean()` apparently causes array mutation, which breaks Zygote's ability to differentiate:
```julia
julia> using Zyg…
-
Howdy,
https://diffeq.sciml.ai/stable/analysis/sensitivity/#Example-continuous-adjoints-on-an-energy-functional
After seeing your response should the latex for the integrand of G(p) be (\sum u_…
-
Hello is it possible to apply checkpointing in backpropagation - I mean recomputing output of some layers instead of saving them in order to reduce memory consumption during backpropagation.
Someth…
-
Evaluating the `sum` of a sparse array is slow. It appears to use an `O(n^2)` algorithm:
```Julia
julia> @time sum(sparse(I(10000)))
0.401920 seconds (6 allocations: 176.281 KiB)
10000
juli…
-
Hey,
when applying `abs2` to a complex CUDA array I get an `ERROR: MethodError: no method matching iterate(::Nothing)`.
I'm using CUDA 3.1.0, Julia 1.6.1 and Zygote 0.6.10.
But I also tried it o…