-
There are some adjoints which repeat calculations in the backwards pass which can be avoided.
https://github.com/FluxML/Zygote.jl/blob/master/src/lib/array.jl#L429-L443
This refactorizes `A'` and…
-
I am getting a strange compilation error after upgrading when compiling a model that uses reduce sum:
```
/Library/Developer/CommandLineTools/usr/bin/../include/c++/v1/memory:3033:32: error: no ma…
-
Hi everyone
I have a small ODE model, which runs fine (albeit slow) with Turing forwarddiff AD. However, I cannot get it to run with reversediff (and neither with Zygote). It throws a MethodError (…
-
In #751 I added some tests to make sure that the adjoints for FFTs preserved the types of the inputs, but I had some trouble with the following test
```julia
x = randn(16,16)
@test typeof(gradien…
-
From JuliaLang: https://discourse.julialang.org/t/using-callback-during-adjoint-method-for-neuralode/70076
I am training a neural ode with a callback function to add a certain value at every integr…
-
Potentially related to https://github.com/SciML/DifferentialEquations.jl/issues/679, but I think it's its own thing. Basically ArrayPartitions, DynamicalODEProblems, and DiffEqSensitivity.jl is an exp…
-
-
Both the gridder and degridder take weights as inputs. This is the most general approach but requires some care on the user's part. Specifically, the user should be made aware that passing the actual …
-
Because adjoints have special scalar dot product, this was a Q&D hack that needs to be fixed. The "right way" is to support infinite `ReshapedArray`.
-