<s>
Eclipse	O
Deeplearning4j	B-Library
is	O
a	O
programming	B-Library
library	I-Library
written	O
in	O
Java	B-Language
for	O
the	O
Java	B-Language
virtual	I-Language
machine	I-Language
(	O
JVM	B-Language
)	O
.	O
</s>
<s>
It	O
is	O
a	O
framework	B-Architecture
with	O
wide	O
support	O
for	O
deep	B-Algorithm
learning	I-Algorithm
algorithms	O
.	O
</s>
<s>
Deeplearning4j	B-Library
includes	O
implementations	O
of	O
the	O
restricted	B-Algorithm
Boltzmann	I-Algorithm
machine	I-Algorithm
,	O
deep	B-Algorithm
belief	I-Algorithm
net	I-Algorithm
,	O
deep	O
autoencoder	O
,	O
stacked	O
denoising	O
autoencoder	O
and	O
recursive	O
neural	O
tensor	O
network	O
,	O
word2vec	B-Algorithm
,	O
doc2vec	O
,	O
and	O
GloVe	B-General_Concept
.	O
</s>
<s>
These	O
algorithms	O
all	O
include	O
distributed	B-Architecture
parallel	B-Operating_System
versions	O
that	O
integrate	O
with	O
Apache	B-Application
Hadoop	I-Application
and	O
Spark	B-Language
.	O
</s>
<s>
Deeplearning4j	B-Library
is	O
open-source	B-Application
software	I-Application
released	O
under	O
Apache	B-Application
License	I-Application
2.0	I-Application
,	O
developed	O
mainly	O
by	O
a	O
machine	O
learning	O
group	O
headquartered	O
in	O
San	O
Francisco	O
.	O
</s>
<s>
It	O
is	O
supported	O
commercially	O
by	O
the	O
startup	O
Skymind	O
,	O
which	O
bundles	O
DL4J	B-Library
,	O
TensorFlow	B-Language
,	O
Keras	B-Algorithm
and	O
other	O
deep	B-Algorithm
learning	I-Algorithm
libraries	O
in	O
an	O
enterprise	O
distribution	O
called	O
the	O
Skymind	O
Intelligence	O
Layer	O
.	O
</s>
<s>
Deeplearning4j	B-Library
was	O
contributed	O
to	O
the	B-Application
Eclipse	I-Application
Foundation	I-Application
in	O
October	O
2017	O
.	O
</s>
<s>
Deeplearning4j	B-Library
relies	O
on	O
the	O
widely	O
used	O
programming	O
language	O
Java	B-Language
,	O
though	O
it	O
is	O
compatible	O
with	O
Clojure	B-Language
and	O
includes	O
a	O
Scala	B-Application
application	B-Application
programming	I-Application
interface	I-Application
(	O
API	B-Application
)	O
.	O
</s>
<s>
It	O
is	O
powered	O
by	O
its	O
own	O
open-source	B-Application
numerical	O
computing	O
library	B-Library
,	O
ND4J	O
,	O
and	O
works	O
with	O
both	O
central	B-General_Concept
processing	I-General_Concept
units	I-General_Concept
(	O
CPUs	O
)	O
and	O
graphics	B-Architecture
processing	I-Architecture
units	I-Architecture
(	O
GPUs	B-Architecture
)	O
.	O
</s>
<s>
Deeplearning4j	B-Library
has	O
been	O
used	O
in	O
several	O
commercial	O
and	O
academic	O
applications	O
.	O
</s>
<s>
The	O
code	O
is	O
hosted	O
on	O
GitHub	B-Application
.	O
</s>
<s>
A	O
support	O
forum	O
is	O
maintained	O
on	O
Gitter	B-Protocol
.	O
</s>
<s>
The	O
framework	B-Architecture
is	O
composable	O
,	O
meaning	O
shallow	O
neural	O
nets	O
such	O
as	O
restricted	B-Algorithm
Boltzmann	I-Algorithm
machines	I-Algorithm
,	O
convolutional	O
nets	O
,	O
autoencoders	O
,	O
and	O
recurrent	O
nets	O
can	O
be	O
added	O
to	O
one	O
another	O
to	O
create	O
deep	O
nets	O
of	O
varying	O
types	O
.	O
</s>
<s>
Training	O
with	O
Deeplearning4j	B-Library
occurs	O
in	O
a	O
cluster	O
.	O
</s>
<s>
Neural	O
nets	O
are	O
trained	O
in	O
parallel	B-Operating_System
via	O
iterative	O
reduce	O
,	O
which	O
works	O
on	O
Hadoop-YARN	O
and	O
on	O
Spark	B-Language
.	O
</s>
<s>
Deeplearning4j	B-Library
also	O
integrates	O
with	O
CUDA	B-Architecture
kernels	O
to	O
conduct	O
pure	O
GPU	B-Architecture
operations	O
,	O
and	O
works	O
with	O
distributed	B-Architecture
GPUs	B-Architecture
.	O
</s>
<s>
Deeplearning4j	B-Library
includes	O
an	O
n-dimensional	O
array	O
class	O
using	O
ND4J	O
that	O
allows	O
scientific	O
computing	O
in	O
Java	B-Language
and	O
Scala	B-Application
,	O
similar	O
to	O
the	O
functions	O
that	O
NumPy	B-Application
provides	O
to	O
Python	B-Language
.	O
</s>
<s>
It	O
's	O
effectively	O
based	O
on	O
a	O
library	B-Library
for	O
linear	B-Language
algebra	I-Language
and	O
matrix	B-Architecture
manipulation	O
in	O
a	O
production	O
environment	O
.	O
</s>
<s>
DataVec	O
vectorizes	O
various	O
file	O
formats	O
and	O
data	O
types	O
using	O
an	O
input/output	B-General_Concept
format	O
system	O
similar	O
to	O
Hadoop	B-Application
's	O
use	O
of	O
MapReduce	O
;	O
that	O
is	O
,	O
it	O
turns	O
various	O
data	O
types	O
into	O
columns	O
of	O
scalars	O
termed	O
vectors	O
.	O
</s>
<s>
Deeplearning4j	B-Library
includes	O
a	O
vector	O
space	O
modeling	O
and	O
topic	O
modeling	O
toolkit	O
,	O
implemented	O
in	O
Java	B-Language
and	O
integrating	O
with	O
parallel	B-Operating_System
GPUs	B-Architecture
for	O
performance	O
.	O
</s>
<s>
Deeplearning4j	B-Library
includes	O
implementations	O
of	O
term	O
frequency	O
–	O
inverse	O
document	O
frequency	O
(	O
tf	O
–	O
idf	O
)	O
,	O
deep	B-Algorithm
learning	I-Algorithm
,	O
and	O
Mikolov	O
's	O
word2vec	B-Algorithm
algorithm	O
,	O
doc2vec	O
,	O
and	O
GloVe	B-General_Concept
,	O
reimplemented	O
and	O
optimized	O
in	O
Java	B-Language
.	O
</s>
<s>
It	O
relies	O
on	O
t-distributed	B-Algorithm
stochastic	I-Algorithm
neighbor	I-Algorithm
embedding	I-Algorithm
(	O
t-SNE	B-Algorithm
)	O
for	O
word-cloud	O
visualizations	O
.	O
</s>
<s>
Real-world	O
use	O
cases	O
for	O
Deeplearning4j	B-Library
include	O
network	O
intrusion	O
detection	O
and	O
cybersecurity	O
,	O
fraud	O
detection	O
for	O
the	O
financial	O
sector	O
,	O
anomaly	O
detection	O
in	O
industries	O
such	O
as	O
manufacturing	O
,	O
recommender	O
systems	O
in	O
e-commerce	O
and	O
advertising	O
,	O
and	O
image	O
recognition	O
.	O
</s>
<s>
Deeplearning4j	B-Library
has	O
integrated	O
with	O
other	O
machine-learning	O
platforms	O
such	O
as	O
RapidMiner	O
,	O
Prediction.io,	O
and	O
Weka	B-Language
.	O
</s>
<s>
Deeplearning4j	B-Library
serves	O
machine-learning	O
models	O
for	O
inference	O
in	O
production	O
using	O
the	O
free	O
developer	O
edition	O
of	O
SKIL	O
,	O
the	O
Skymind	O
Intelligence	O
Layer	O
.	O
</s>
<s>
What	O
a	O
Web	O
server	O
is	O
to	O
the	O
Internet	O
,	O
a	O
model	O
server	O
is	O
to	O
AI	B-Application
.	O
</s>
<s>
The	O
SKIL	O
model	O
server	O
is	O
able	O
to	O
import	O
models	O
from	O
Python	B-Language
frameworks	O
such	O
as	O
Tensorflow	B-Language
,	O
Keras	B-Algorithm
,	O
Theano	O
and	O
CNTK	O
,	O
overcoming	O
a	O
major	O
barrier	O
in	O
deploying	O
deep	B-Algorithm
learning	I-Algorithm
models	O
.	O
</s>
<s>
Deeplearning4j	B-Library
is	O
as	O
fast	O
as	O
Caffe	O
for	O
non-trivial	O
image	O
recognition	O
tasks	O
using	O
multiple	O
GPUs	B-Architecture
.	O
</s>
<s>
For	O
programmers	O
unfamiliar	O
with	O
HPC	O
on	O
the	O
JVM	B-Language
,	O
there	O
are	O
several	O
parameters	O
that	O
must	O
be	O
adjusted	O
to	O
optimize	O
neural	O
network	O
training	O
time	O
.	O
</s>
<s>
Together	O
,	O
these	O
optimizations	O
can	O
lead	O
to	O
a	O
10x	O
acceleration	O
in	O
performance	O
with	O
Deeplearning4j	B-Library
.	O
</s>
<s>
Deeplearning4j	B-Library
can	O
be	O
used	O
via	O
multiple	O
API	B-Application
languages	O
including	O
Java	B-Language
,	O
Scala	B-Application
,	O
Python	B-Language
,	O
Clojure	B-Language
and	O
Kotlin	O
.	O
</s>
<s>
Its	O
Scala	B-Application
API	B-Application
is	O
called	O
ScalNet	O
.	O
</s>
<s>
Keras	B-Algorithm
serves	O
as	O
its	O
Python	B-Language
API	B-Application
.	O
</s>
<s>
And	O
its	O
Clojure	B-Language
wrapper	O
is	O
known	O
as	O
DL4CLJ	O
.	O
</s>
<s>
The	O
core	O
languages	O
performing	O
the	O
large-scale	O
mathematical	O
operations	O
necessary	O
for	O
deep	B-Algorithm
learning	I-Algorithm
are	O
C	B-Language
,	O
C++	B-Language
and	O
CUDA	B-Architecture
C	B-Language
.	O
</s>
<s>
Tensorflow	B-Language
,	O
Keras	B-Algorithm
and	O
Deeplearning4j	B-Library
work	O
together	O
.	O
</s>
<s>
Deeplearning4j	B-Library
can	O
import	O
models	O
from	O
Tensorflow	B-Language
and	O
other	O
Python	B-Language
frameworks	O
if	O
they	O
have	O
been	O
created	O
with	O
Keras	B-Algorithm
.	O
</s>
