<s>
TensorFlow	B-Language
is	O
a	O
free	B-License
and	I-License
open-source	I-License
software	I-License
library	B-Library
for	O
machine	O
learning	O
and	O
artificial	B-Application
intelligence	I-Application
.	I-Application
</s>
<s>
It	O
can	O
be	O
used	O
across	O
a	O
range	O
of	O
tasks	O
but	O
has	O
a	O
particular	O
focus	O
on	O
training	B-General_Concept
and	O
inference	O
of	O
deep	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
TensorFlow	B-Language
was	O
developed	O
by	O
the	B-Application
Google	I-Application
Brain	O
team	O
for	O
internal	O
Google	B-Application
use	O
in	O
research	O
and	O
production	O
.	O
</s>
<s>
The	O
initial	O
version	O
was	O
released	O
under	O
the	O
Apache	B-Application
License	I-Application
2.0	I-Application
in	O
2015	O
.	O
</s>
<s>
Google	B-Application
released	O
the	O
updated	O
version	O
of	O
TensorFlow	B-Language
,	O
named	O
TensorFlow	B-Language
2.0	O
,	O
in	O
September	O
2019	O
.	O
</s>
<s>
TensorFlow	B-Language
can	O
be	O
used	O
in	O
a	O
wide	O
variety	O
of	O
programming	O
languages	O
,	O
including	O
Python	B-Language
,	O
JavaScript	B-Language
,	O
C++	B-Language
,	O
and	O
Java	B-Language
.	O
</s>
<s>
Starting	O
in	O
2011	O
,	O
Google	B-Application
Brain	I-Application
built	O
DistBelief	O
as	O
a	O
proprietary	B-Application
machine	O
learning	O
system	O
based	O
on	O
deep	B-Algorithm
learning	I-Algorithm
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
Google	B-Application
assigned	O
multiple	O
computer	O
scientists	O
,	O
including	O
Jeff	O
Dean	O
,	O
to	O
simplify	O
and	O
refactor	O
the	O
codebase	O
of	O
DistBelief	O
into	O
a	O
faster	O
,	O
more	O
robust	O
application-grade	O
library	B-Library
,	O
which	O
became	O
TensorFlow	B-Language
.	O
</s>
<s>
In	O
2009	O
,	O
the	O
team	O
,	O
led	O
by	O
Geoffrey	O
Hinton	O
,	O
had	O
implemented	O
generalized	O
backpropagation	B-Algorithm
and	O
other	O
improvements	O
which	O
allowed	O
generation	O
of	O
neural	B-Architecture
networks	I-Architecture
with	O
substantially	O
higher	O
accuracy	O
,	O
for	O
instance	O
a	O
25%	O
reduction	O
in	O
errors	O
in	O
speech	B-Application
recognition	I-Application
.	O
</s>
<s>
TensorFlow	B-Language
is	O
Google	B-Application
Brain	I-Application
's	O
second-generation	O
system	O
.	O
</s>
<s>
While	O
the	O
reference	O
implementation	O
runs	O
on	O
single	O
devices	O
,	O
TensorFlow	B-Language
can	O
run	O
on	O
multiple	O
CPUs	B-General_Concept
and	O
GPUs	B-Architecture
(	O
with	O
optional	O
CUDA	B-Architecture
and	O
SYCL	B-Language
extensions	O
for	O
general-purpose	B-Architecture
computing	I-Architecture
on	I-Architecture
graphics	I-Architecture
processing	I-Architecture
units	I-Architecture
)	O
.	O
</s>
<s>
TensorFlow	B-Language
is	O
available	O
on	O
64-bit	O
Linux	B-Application
,	O
macOS	B-Application
,	O
Windows	O
,	O
and	O
mobile	O
computing	O
platforms	O
including	O
Android	B-Application
and	O
iOS	B-Application
.	O
</s>
<s>
Its	O
flexible	O
architecture	O
allows	O
for	O
the	O
easy	O
deployment	O
of	O
computation	O
across	O
a	O
variety	O
of	O
platforms	O
(	O
CPUs	B-General_Concept
,	O
GPUs	B-Architecture
,	O
TPUs	O
)	O
,	O
and	O
from	O
desktops	O
to	O
clusters	O
of	O
servers	O
to	O
mobile	O
and	O
edge	O
devices	O
.	O
</s>
<s>
TensorFlow	B-Language
computations	O
are	O
expressed	O
as	O
stateful	B-Application
dataflow	B-Application
graphs	O
.	O
</s>
<s>
The	O
name	O
TensorFlow	B-Language
derives	O
from	O
the	O
operations	O
that	O
such	O
neural	B-Architecture
networks	I-Architecture
perform	O
on	O
multidimensional	O
data	O
arrays	O
,	O
which	O
are	O
referred	O
to	O
as	O
tensors	B-Device
.	O
</s>
<s>
During	O
the	B-Application
Google	I-Application
I/O	O
Conference	O
in	O
June	O
2016	O
,	O
Jeff	O
Dean	O
stated	O
that	O
1,500	O
repositories	O
on	O
GitHub	B-Application
mentioned	O
TensorFlow	B-Language
,	O
of	O
which	O
only	O
5	O
were	O
from	O
Google	B-Application
.	I-Application
</s>
<s>
In	O
December	O
2017	O
,	O
developers	O
from	O
Google	B-Application
,	O
Cisco	O
,	O
RedHat	O
,	O
CoreOS	O
,	O
and	O
CaiCloud	O
introduced	O
Kubeflow	B-General_Concept
at	O
a	O
conference	O
.	O
</s>
<s>
Kubeflow	B-General_Concept
allows	O
operation	O
and	O
deployment	O
of	O
TensorFlow	B-Language
on	O
Kubernetes	B-General_Concept
.	O
</s>
<s>
In	O
March	O
2018	O
,	O
Google	B-Application
announced	O
TensorFlow.js	O
version	O
1.0	O
for	O
machine	O
learning	O
in	O
JavaScript	B-Language
.	O
</s>
<s>
In	O
Jan	O
2019	O
,	O
Google	B-Application
announced	O
TensorFlow	B-Language
2.0	O
.	O
</s>
<s>
In	O
May	O
2019	O
,	O
Google	B-Application
announced	O
TensorFlow	B-Language
Graphics	O
for	O
deep	B-Algorithm
learning	I-Algorithm
in	O
computer	O
graphics	O
.	O
</s>
<s>
In	O
May	O
2016	O
,	O
Google	B-Application
announced	O
its	O
Tensor	B-Device
processing	I-Device
unit	I-Device
(	O
TPU	B-Device
)	O
,	O
an	O
application-specific	O
integrated	O
circuit	O
(	O
ASIC	O
,	O
a	O
hardware	O
chip	O
)	O
built	O
specifically	O
for	O
machine	O
learning	O
and	O
tailored	O
for	O
TensorFlow	B-Language
.	O
</s>
<s>
A	O
TPU	B-Device
is	O
a	O
programmable	O
AI	B-General_Concept
accelerator	I-General_Concept
designed	O
to	O
provide	O
high	O
throughput	O
of	O
low-precision	O
arithmetic	O
(	O
e.g.	O
,	O
8-bit	O
)	O
,	O
and	O
oriented	O
toward	O
using	O
or	O
running	O
models	O
rather	O
than	O
training	B-General_Concept
them	O
.	O
</s>
<s>
Google	B-Application
announced	O
they	O
had	O
been	O
running	O
TPUs	O
inside	O
their	O
data	O
centers	O
for	O
more	O
than	O
a	O
year	O
,	O
and	O
had	O
found	O
them	O
to	O
deliver	O
an	O
order	O
of	O
magnitude	O
better-optimized	O
performance	O
per	O
watt	O
for	O
machine	O
learning	O
.	O
</s>
<s>
In	O
May	O
2017	O
,	O
Google	B-Application
announced	O
the	O
second-generation	O
,	O
as	O
well	O
as	O
the	O
availability	O
of	O
the	O
TPUs	O
in	O
Google	B-General_Concept
Compute	I-General_Concept
Engine	I-General_Concept
.	O
</s>
<s>
In	O
May	O
2018	O
,	O
Google	B-Application
announced	O
the	O
third-generation	O
TPUs	O
delivering	O
up	O
to	O
420	O
teraflops	O
of	O
performance	O
and	O
128	O
GB	O
high	O
bandwidth	O
memory	O
(	O
HBM	O
)	O
.	O
</s>
<s>
Cloud	O
TPU	B-Device
v3	O
Pods	O
offer	O
100+	O
petaflops	O
of	O
performance	O
and	O
32	O
TB	O
HBM	O
.	O
</s>
<s>
In	O
February	O
2018	O
,	O
Google	B-Application
announced	O
that	O
they	O
were	O
making	O
TPUs	O
available	O
in	O
beta	O
on	O
the	B-Application
Google	I-Application
Cloud	O
Platform	O
.	O
</s>
<s>
In	O
July	O
2018	O
,	O
the	O
Edge	O
TPU	B-Device
was	O
announced	O
.	O
</s>
<s>
Edge	O
TPU	B-Device
is	O
Google	B-Application
's	I-Application
purpose-built	O
ASIC	O
chip	O
designed	O
to	O
run	O
TensorFlow	B-Language
Lite	O
machine	O
learning	O
(	O
ML	O
)	O
models	O
on	O
small	O
client	O
computing	O
devices	O
such	O
as	O
smartphones	O
known	O
as	O
edge	B-Device
computing	I-Device
.	O
</s>
<s>
In	O
May	O
2017	O
,	O
Google	B-Application
announced	O
a	O
software	O
stack	O
specifically	O
for	O
mobile	O
development	O
,	O
TensorFlow	B-Language
Lite	O
.	O
</s>
<s>
In	O
January	O
2019	O
,	O
the	O
TensorFlow	B-Language
team	O
released	O
a	O
developer	O
preview	O
of	O
the	O
mobile	O
GPU	B-Architecture
inference	O
engine	O
with	O
OpenGL	O
ES	O
3.1	O
Compute	O
Shaders	O
on	O
Android	B-Application
devices	O
and	O
Metal	O
Compute	O
Shaders	O
on	O
iOS	B-Application
devices	O
.	O
</s>
<s>
In	O
May	O
2019	O
,	O
Google	B-Application
announced	O
that	O
their	O
TensorFlow	B-Language
Lite	O
Micro	O
(	O
also	O
known	O
as	O
TensorFlow	B-Language
Lite	O
for	O
Microcontrollers	O
)	O
and	O
ARM	O
's	O
uTensor	O
would	O
be	O
merging	O
.	O
</s>
<s>
In	O
October	O
2017	O
,	O
Google	B-Application
released	O
the	B-Application
Google	I-Application
Pixel	B-Device
2	I-Device
which	O
featured	O
their	O
Pixel	B-Device
Visual	I-Device
Core	I-Device
(	O
PVC	O
)	O
,	O
a	O
fully	O
programmable	O
image	B-General_Concept
,	O
vision	B-General_Concept
and	O
AI	B-General_Concept
processor	O
for	O
mobile	O
devices	O
.	O
</s>
<s>
The	O
PVC	O
supports	O
TensorFlow	B-Language
for	O
machine	O
learning	O
(	O
and	O
Halide	B-Operating_System
for	O
image	B-General_Concept
processing	O
)	O
.	O
</s>
<s>
As	O
TensorFlow	B-Language
's	O
market	O
share	O
among	O
research	O
papers	O
was	O
declining	O
to	O
the	O
advantage	O
of	O
PyTorch	B-Algorithm
,	O
the	O
TensorFlow	B-Language
Team	O
announced	O
a	O
release	O
of	O
a	O
new	O
major	O
version	O
of	O
the	O
library	B-Library
in	O
September	O
2019	O
.	O
</s>
<s>
TensorFlow	B-Language
2.0	O
introduced	O
many	O
changes	O
,	O
the	O
most	O
significant	O
being	O
TensorFlow	B-Language
eager	O
,	O
which	O
changed	O
the	O
automatic	B-Algorithm
differentiation	I-Algorithm
scheme	O
from	O
the	O
static	O
computational	O
graph	O
,	O
to	O
the	O
"	O
Define-by-Run	O
"	O
scheme	O
originally	O
made	O
popular	O
by	O
Chainer	B-Algorithm
and	O
later	O
PyTorch	B-Algorithm
.	O
</s>
<s>
Other	O
major	O
changes	O
included	O
removal	O
of	O
old	O
libraries	O
,	O
cross-compatibility	O
between	O
trained	O
models	O
on	O
different	O
versions	O
of	O
TensorFlow	B-Language
,	O
and	O
significant	O
improvements	O
to	O
the	O
performance	O
on	O
GPU	B-Architecture
.	O
</s>
<s>
AutoDifferentiation	B-Algorithm
is	O
the	O
process	O
of	O
automatically	O
calculating	O
the	O
gradient	O
vector	O
of	O
a	O
model	O
with	O
respect	O
to	O
each	O
of	O
its	O
parameters	O
.	O
</s>
<s>
With	O
this	O
feature	O
,	O
TensorFlow	B-Language
can	O
automatically	O
compute	O
the	O
gradients	O
for	O
the	O
parameters	O
in	O
a	O
model	O
,	O
which	O
is	O
useful	O
to	O
algorithms	O
such	O
as	O
backpropagation	B-Algorithm
which	O
require	O
gradients	O
to	O
optimize	O
performance	O
.	O
</s>
<s>
To	O
do	O
so	O
,	O
the	O
framework	O
must	O
keep	O
track	O
of	O
the	O
order	O
of	O
operations	O
done	O
to	O
the	O
input	O
Tensors	B-Device
in	O
a	O
model	O
,	O
and	O
then	O
compute	O
the	O
gradients	O
with	O
respect	O
to	O
the	O
appropriate	O
parameters	O
.	O
</s>
<s>
TensorFlow	B-Language
includes	O
an	O
“	O
eager	O
execution	O
”	O
mode	O
,	O
which	O
means	O
that	O
operations	O
are	O
evaluated	O
immediately	O
as	O
opposed	O
to	O
being	O
added	O
to	O
a	O
computational	O
graph	O
which	O
is	O
executed	O
later	O
.	O
</s>
<s>
In	O
both	O
eager	O
and	O
graph	O
executions	O
,	O
TensorFlow	B-Language
provides	O
an	O
API	O
for	O
distributing	O
computation	O
across	O
multiple	O
devices	O
with	O
various	O
distribution	O
strategies	O
.	O
</s>
<s>
This	O
distributed	B-Architecture
computing	I-Architecture
can	O
often	O
speed	O
up	O
the	O
execution	O
of	O
training	B-General_Concept
and	O
evaluating	O
of	O
TensorFlow	B-Language
models	O
and	O
is	O
a	O
common	O
practice	O
in	O
the	O
field	O
of	O
AI	B-General_Concept
.	O
</s>
<s>
To	O
train	O
and	O
assess	O
models	O
,	O
TensorFlow	B-Language
provides	O
a	O
set	O
of	O
loss	O
functions	O
(	O
also	O
known	O
as	O
cost	O
functions	O
)	O
.	O
</s>
<s>
Some	O
popular	O
examples	O
include	O
mean	B-Algorithm
squared	I-Algorithm
error	I-Algorithm
(	O
MSE	O
)	O
and	O
binary	O
cross	O
entropy	O
(	O
BCE	O
)	O
.	O
</s>
<s>
These	O
loss	O
functions	O
compute	O
the	O
“	O
error	O
”	O
or	O
“	O
difference	O
”	O
between	O
a	O
model	O
's	O
output	O
and	O
the	O
expected	O
output	O
(	O
more	O
broadly	O
,	O
the	O
difference	O
between	O
two	O
tensors	B-Device
)	O
.	O
</s>
<s>
In	O
order	O
to	O
assess	O
the	O
performance	O
of	O
machine	O
learning	O
models	O
,	O
TensorFlow	B-Language
gives	O
API	O
access	O
to	O
commonly	O
used	O
metrics	O
.	O
</s>
<s>
TensorFlow.nn	O
is	O
a	O
module	O
for	O
executing	O
primitive	O
neural	B-Architecture
network	I-Architecture
operations	O
on	O
models	O
.	O
</s>
<s>
Some	O
of	O
these	O
operations	O
include	O
variations	O
of	O
convolutions	B-Architecture
(	O
1/2/3D	O
,	O
Atrous	O
,	O
depthwise	O
)	O
,	O
activation	B-Algorithm
functions	I-Algorithm
(	O
Softmax	B-Algorithm
,	O
RELU	B-Algorithm
,	O
GELU	O
,	O
Sigmoid	B-Algorithm
,	O
etc	O
.	O
)	O
</s>
<s>
and	O
their	O
variations	O
,	O
and	O
other	O
Tensor	B-Device
operations	O
(	O
max-pooling	O
,	O
bias-add	O
,	O
etc	O
.	O
</s>
<s>
TensorFlow	B-Language
offers	O
a	O
set	O
of	O
optimizers	O
for	O
training	B-General_Concept
neural	B-Architecture
networks	I-Architecture
,	O
including	O
ADAM	O
,	O
ADAGRAD	B-Algorithm
,	O
and	O
Stochastic	B-Algorithm
Gradient	I-Algorithm
Descent	I-Algorithm
(	O
SGD	O
)	O
.	O
</s>
<s>
When	O
training	B-General_Concept
a	O
model	O
,	O
different	O
optimizers	O
offer	O
different	O
modes	O
of	O
parameter	O
tuning	O
,	O
often	O
affecting	O
a	O
model	O
's	O
convergence	O
and	O
performance	O
.	O
</s>
<s>
TensorFlow	B-Language
serves	O
as	O
a	O
core	O
platform	O
and	O
library	B-Library
for	O
machine	O
learning	O
.	O
</s>
<s>
TensorFlow	B-Language
's	O
APIs	O
use	O
Keras	B-Algorithm
to	O
allow	O
users	O
to	O
make	O
their	O
own	O
machine	O
learning	O
models	O
.	O
</s>
<s>
In	O
addition	O
to	O
building	O
and	O
training	B-General_Concept
their	O
model	O
,	O
TensorFlow	B-Language
can	O
also	O
help	O
load	O
the	O
data	O
to	O
train	O
the	O
model	O
,	O
and	O
deploy	O
it	O
using	O
TensorFlow	B-Language
Serving	O
.	O
</s>
<s>
TensorFlow	B-Language
provides	O
a	O
stable	O
Python	B-Language
API	O
,	O
as	O
well	O
as	O
APIs	O
without	O
backwards	O
compatibility	O
guarantee	O
for	O
Javascript	B-Language
,	O
C++	B-Language
,	O
and	O
Java	B-Language
.	O
</s>
<s>
Third-party	O
language	O
binding	O
packages	O
are	O
also	O
available	O
for	O
C#	B-Application
,	O
Haskell	B-Language
,	O
Julia	B-Application
,	O
MATLAB	B-Language
,	O
Object	B-Language
Pascal	I-Language
,	O
R	B-Language
,	O
Scala	B-Application
,	O
Rust	B-Application
,	O
OCaml	B-Language
,	O
and	O
Crystal	B-Application
.	O
</s>
<s>
Bindings	O
that	O
are	O
now	O
archived	O
and	O
unsupported	O
include	O
Go	B-Application
and	O
Swift	B-Application
.	O
</s>
<s>
TensorFlow	B-Language
also	O
has	O
a	O
library	B-Library
for	O
machine	O
learning	O
in	O
JavaScript	B-Language
.	O
</s>
<s>
Using	O
the	O
provided	O
JavaScript	B-Language
APIs	O
,	O
TensorFlow.js	O
allows	O
users	O
to	O
use	O
either	O
Tensorflow.js	O
models	O
or	O
converted	O
models	O
from	O
TensorFlow	B-Language
or	O
TFLite	O
,	O
retrain	O
the	O
given	O
models	O
,	O
and	O
run	O
on	O
the	O
web	O
.	O
</s>
<s>
TensorFlow	B-Language
Lite	O
has	O
APIs	O
for	O
mobile	O
apps	O
or	O
embedded	O
devices	O
to	O
generate	O
and	O
deploy	O
TensorFlow	B-Language
models	O
.	O
</s>
<s>
TensorFlow	B-Language
Lite	O
uses	O
FlatBuffers	B-Protocol
as	O
the	O
data	O
serialization	O
format	O
for	O
network	O
models	O
,	O
eschewing	O
the	O
Protocol	B-Protocol
Buffers	I-Protocol
format	O
used	O
by	O
standard	O
TensorFlow	B-Language
models	O
.	O
</s>
<s>
TensorFlow	B-Language
Extended	O
(	O
abbrev	O
.	O
</s>
<s>
Components	O
include	O
loading	O
,	O
validating	O
,	O
and	O
transforming	O
data	O
,	O
tuning	O
,	O
training	B-General_Concept
,	O
and	O
evaluating	O
the	O
machine	O
learning	O
model	O
,	O
and	O
pushing	O
the	O
model	O
itself	O
into	O
production	O
.	O
</s>
<s>
Numpy	B-Application
is	O
one	O
of	O
the	O
most	O
popular	O
Python	B-Language
data	O
libraries	O
,	O
and	O
TensorFlow	B-Language
offers	O
integration	O
and	O
compatibility	O
with	O
its	O
data	O
structures	O
.	O
</s>
<s>
Numpy	B-Application
NDarrays	O
,	O
the	O
library	B-Library
's	O
native	O
datatype	O
,	O
are	O
automatically	O
converted	O
to	O
TensorFlow	B-Language
Tensors	B-Device
in	O
TF	O
operations	O
;	O
the	O
same	O
is	O
also	O
true	O
vice	O
versa	O
.	O
</s>
<s>
Moreover	O
,	O
the	O
integration	O
extends	O
to	O
memory	O
optimization	O
by	O
having	O
TF	O
Tensors	B-Device
share	O
the	O
underlying	O
memory	O
representations	O
of	O
Numpy	B-Application
NDarrays	O
whenever	O
possible	O
.	O
</s>
<s>
TensorFlow	B-Language
also	O
offers	O
a	O
variety	O
of	O
libraries	O
and	O
extensions	O
to	O
advance	O
and	O
extend	O
the	O
models	O
and	O
methods	O
used	O
.	O
</s>
<s>
For	O
example	O
,	O
TensorFlow	B-Language
Recommenders	O
and	O
TensorFlow	B-Language
Graphics	O
are	O
libraries	O
for	O
their	O
respective	O
functionalities	O
in	O
recommendation	O
systems	O
and	O
graphics	O
,	O
TensorFlow	B-Language
Federated	O
provides	O
a	O
framework	O
for	O
decentralized	O
data	O
,	O
and	O
TensorFlow	B-Language
Cloud	O
allows	O
users	O
to	O
directly	O
interact	O
with	O
Google	B-Application
Cloud	O
to	O
integrate	O
their	O
local	O
code	O
to	O
Google	B-Application
Cloud	O
.	O
</s>
<s>
Other	O
add-ons	O
,	O
libraries	O
,	O
and	O
frameworks	O
include	O
TensorFlow	B-Language
Model	O
Optimization	O
,	O
TensorFlow	B-Language
Probability	O
,	O
TensorFlow	B-Language
Quantum	O
,	O
and	O
TensorFlow	B-Language
Decision	O
Forests	O
.	O
</s>
<s>
Google	B-Application
also	O
released	O
Colaboratory	O
,	O
a	O
TensorFlow	B-Language
Jupyter	O
notebook	O
environment	O
that	O
does	O
not	O
require	O
any	O
setup	O
.	O
</s>
<s>
It	O
runs	O
on	O
Google	B-Application
Cloud	O
and	O
allows	O
users	O
free	O
access	O
to	O
GPUs	B-Architecture
and	O
the	O
ability	O
to	O
store	O
and	O
share	O
notebooks	O
on	O
Google	B-Application
Drive	O
.	O
</s>
<s>
Google	B-Application
JAX	I-Application
is	O
a	O
machine	O
learning	O
framework	O
for	O
transforming	O
numerical	O
functions	O
.	O
</s>
<s>
It	O
is	O
described	O
as	O
bringing	O
together	O
a	O
modified	O
version	O
of	O
(	O
automatic	O
obtaining	O
of	O
the	O
gradient	O
function	O
through	O
differentiation	O
of	O
a	O
function	O
)	O
and	O
TensorFlow	B-Language
's	O
(	O
Accelerated	O
Linear	O
Algebra	O
)	O
.	O
</s>
<s>
It	O
is	O
designed	O
to	O
follow	O
the	O
structure	O
and	O
workflow	O
of	O
NumPy	B-Application
as	O
closely	O
as	O
possible	O
and	O
works	O
with	O
TensorFlow	B-Language
as	O
well	O
as	O
other	O
frameworks	O
such	O
as	O
PyTorch	B-Algorithm
.	O
</s>
<s>
GE	O
Healthcare	O
used	O
TensorFlow	B-Language
to	O
increase	O
the	O
speed	O
and	O
accuracy	O
of	O
MRIs	O
in	O
identifying	O
specific	O
body	O
parts	O
.	O
</s>
<s>
Google	B-Application
used	O
TensorFlow	B-Language
to	O
create	O
DermAssist	O
,	O
a	O
free	O
mobile	O
application	O
that	O
allows	O
users	O
to	O
take	O
pictures	O
of	O
their	O
skin	O
and	O
identify	O
potential	O
health	O
complications	O
.	O
</s>
<s>
Sinovation	O
Ventures	O
used	O
TensorFlow	B-Language
to	O
identify	O
and	O
classify	O
eye	O
diseases	O
from	O
optical	B-Application
coherence	I-Application
tomography	I-Application
(	O
OCT	O
)	O
scans	O
.	O
</s>
<s>
Twitter	B-Application
implemented	O
TensorFlow	B-Language
to	O
rank	O
tweets	O
by	O
importance	O
for	O
a	O
given	O
user	O
,	O
and	O
changed	O
their	O
platform	O
to	O
show	O
tweets	O
in	O
order	O
of	O
this	O
ranking	O
.	O
</s>
<s>
The	O
photo	O
sharing	O
app	O
VSCO	O
used	O
TensorFlow	B-Language
to	O
help	O
suggest	O
custom	O
filters	O
for	O
photos	O
.	O
</s>
<s>
Google	B-Application
officially	O
released	O
RankBrain	B-Application
on	O
October	O
26	O
,	O
2015	O
,	O
backed	O
by	O
TensorFlow	B-Language
.	O
</s>
<s>
InSpace	O
,	O
a	O
virtual	O
learning	O
platform	O
,	O
used	O
TensorFlow	B-Language
to	O
filter	O
out	O
toxic	O
chat	O
messages	O
in	O
classrooms	O
.	O
</s>
<s>
Liulishuo	O
,	O
an	O
online	O
English	O
learning	O
platform	O
,	O
utilized	O
TensorFlow	B-Language
to	O
create	O
an	O
adaptive	O
curriculum	O
for	O
each	O
student	O
.	O
</s>
<s>
TensorFlow	B-Language
was	O
used	O
to	O
accurately	O
assess	O
a	O
student	O
's	O
current	O
abilities	O
,	O
and	O
also	O
helped	O
decide	O
the	O
best	O
future	O
content	O
to	O
show	O
based	O
on	O
those	O
capabilities	O
.	O
</s>
<s>
The	O
e-commerce	O
platform	O
Carousell	O
used	O
TensorFlow	B-Language
to	O
provide	O
personalized	O
recommendations	O
for	O
customers	O
.	O
</s>
<s>
The	O
cosmetics	O
company	O
ModiFace	O
used	O
TensorFlow	B-Language
to	O
create	O
an	O
augmented	O
reality	O
experience	O
for	O
customers	O
to	O
test	O
various	O
shades	O
of	O
make-up	O
on	O
their	O
face	O
.	O
</s>
<s>
TensorFlow	B-Language
is	O
the	O
foundation	O
for	O
the	O
automated	O
image-captioning	O
software	O
DeepDream	B-Application
.	O
</s>
