<s>
Flux	B-Application
is	O
an	O
open-source	B-License
machine-learning	O
software	B-Library
library	I-Library
and	O
ecosystem	O
written	O
in	O
Julia	B-Application
.	O
</s>
<s>
It	O
has	O
a	O
layer-stacking-based	O
interface	O
for	O
simpler	O
models	O
,	O
and	O
has	O
a	O
strong	O
support	O
on	O
interoperability	O
with	O
other	O
Julia	B-Application
packages	O
instead	O
of	O
a	O
monolithic	O
design	O
.	O
</s>
<s>
For	O
example	O
,	O
GPU	O
support	O
is	O
implemented	O
transparently	O
by	O
CuArrays.jl	O
This	O
is	O
in	O
contrast	O
to	O
some	O
other	O
machine	O
learning	O
frameworks	O
which	O
are	O
implemented	O
in	O
other	O
languages	O
with	O
Julia	B-Application
bindings	O
,	O
such	O
as	O
TensorFlow.jl	B-Language
,	O
and	O
thus	O
are	O
more	O
limited	O
by	O
the	O
functionality	O
present	O
in	O
the	O
underlying	O
implementation	O
,	O
which	O
is	O
often	O
in	O
C	O
or	O
C++	O
.	O
</s>
<s>
Flux	B-Application
joined	O
NumFOCUS	O
as	O
an	O
affiliated	O
project	O
in	O
December	O
of	O
2021	O
.	O
</s>
<s>
Flux	B-Application
's	O
focus	O
on	O
interoperability	O
has	O
enabled	O
,	O
for	O
example	O
,	O
support	O
for	O
Neural	O
Differential	O
Equations	O
,	O
by	O
fusing	O
Flux.jl	O
and	O
DifferentialEquations.jl	O
into	O
DiffEqFlux.jl	O
.	O
</s>
<s>
Flux	B-Application
supports	O
recurrent	O
and	O
convolutional	O
networks	O
.	O
</s>
<s>
It	O
is	O
also	O
capable	O
of	O
differentiable	B-Application
programming	I-Application
through	O
its	O
source-to-source	O
automatic	B-Algorithm
differentiation	I-Algorithm
package	O
,	O
Zygote.jl	O
.	O
</s>
<s>
Julia	B-Application
is	O
a	O
popular	O
language	O
in	O
machine-learning	O
and	O
Flux.jl	O
is	O
its	O
most	O
highly	O
regarded	O
machine-learning	O
repository	O
.	O
</s>
<s>
A	O
demonstration	O
compiling	O
Julia	B-Application
code	O
to	O
run	O
in	O
Google	O
's	O
tensor	B-Device
processing	I-Device
unit	I-Device
(	O
TPU	O
)	O
received	O
praise	O
from	O
Google	B-Application
Brain	I-Application
AI	B-Application
lead	O
Jeff	O
Dean	O
.	O
</s>
<s>
Flux	B-Application
has	O
been	O
used	O
as	O
a	O
framework	O
to	O
build	O
neural	O
networks	O
that	O
work	O
with	O
homomorphic	O
encrypted	O
data	O
without	O
ever	O
decrypting	O
it	O
.	O
</s>
<s>
This	O
kind	O
of	O
application	O
is	O
envisioned	O
to	O
be	O
central	O
for	O
privacy	O
to	O
future	O
API	B-Application
using	O
machine-learning	O
models	O
.	O
</s>
<s>
Flux.jl	O
is	O
an	O
intermediate	B-Application
representation	I-Application
for	O
running	O
high	O
level	O
programs	O
on	O
CUDA	B-Architecture
hardware	O
.	O
</s>
<s>
It	O
was	O
the	O
predecessor	O
to	O
CUDAnative.jl	O
which	O
is	O
also	O
a	O
GPU	B-Architecture
programming	I-Architecture
language	O
.	O
</s>
