<s>
SqueezeNet	B-Application
is	O
the	O
name	O
of	O
a	O
deep	O
neural	O
network	O
for	O
computer	B-Application
vision	I-Application
that	O
was	O
released	O
in	O
2016	O
.	O
</s>
<s>
SqueezeNet	B-Application
was	O
developed	O
by	O
researchers	O
at	O
DeepScale	O
,	O
University	O
of	O
California	O
,	O
Berkeley	O
,	O
and	O
Stanford	O
University	O
.	O
</s>
<s>
In	O
designing	O
SqueezeNet	B-Application
,	O
the	O
authors	O
 '	O
goal	O
was	O
to	O
create	O
a	O
smaller	O
neural	O
network	O
with	O
fewer	O
parameters	O
that	O
can	O
more	O
easily	O
fit	O
into	O
computer	O
memory	O
and	O
can	O
more	O
easily	O
be	O
transmitted	O
over	O
a	O
computer	O
network	O
.	O
</s>
<s>
SqueezeNet	B-Application
was	O
originally	O
released	O
on	O
February	O
22	O
,	O
2016	O
.	O
</s>
<s>
This	O
original	O
version	O
of	O
SqueezeNet	B-Application
was	O
implemented	O
on	O
top	O
of	O
the	O
Caffe	B-Algorithm
deep	O
learning	O
software	O
framework	O
.	O
</s>
<s>
Shortly	O
thereafter	O
,	O
the	O
open-source	O
research	O
community	O
ported	O
SqueezeNet	B-Application
to	O
a	O
number	O
of	O
other	O
deep	O
learning	O
frameworks	O
.	O
</s>
<s>
On	O
February	O
26	O
,	O
2016	O
,	O
Eddie	O
Bell	O
released	O
a	O
port	O
of	O
SqueezeNet	B-Application
for	O
the	O
Chainer	O
deep	O
learning	O
framework	O
.	O
</s>
<s>
On	O
March	O
2	O
,	O
2016	O
,	O
Guo	O
Haria	O
released	O
a	O
port	O
of	O
SqueezeNet	B-Application
for	O
the	O
Apache	B-Algorithm
MXNet	I-Algorithm
framework	O
.	O
</s>
<s>
On	O
June	O
3	O
,	O
2016	O
,	O
Tammy	O
Yang	O
released	O
a	O
port	O
of	O
SqueezeNet	B-Application
for	O
the	O
Keras	B-Algorithm
framework	O
.	O
</s>
<s>
In	O
2017	O
,	O
companies	O
including	O
Baidu	B-Application
,	O
Xilinx	O
,	O
Imagination	O
Technologies	O
,	O
and	O
Synopsys	O
demonstrated	O
SqueezeNet	B-Application
running	O
on	O
low-power	O
processing	O
platforms	O
such	O
as	O
smartphones	B-Application
,	O
FPGAs	B-Architecture
,	O
and	O
custom	O
processors	O
.	O
</s>
<s>
As	O
of	O
2018	O
,	O
SqueezeNet	B-Application
ships	O
"	O
natively	O
"	O
as	O
part	O
of	O
the	O
source	O
code	O
of	O
a	O
number	O
of	O
deep	O
learning	O
frameworks	O
such	O
as	O
PyTorch	B-Algorithm
,	O
Apache	B-Algorithm
MXNet	I-Algorithm
,	O
and	O
Apple	O
CoreML	O
.	O
</s>
<s>
In	O
addition	O
,	O
3rd	O
party	O
developers	O
have	O
created	O
implementations	O
of	O
SqueezeNet	B-Application
that	O
are	O
compatible	O
with	O
frameworks	O
such	O
as	O
TensorFlow	B-Language
.	O
</s>
<s>
Below	O
is	O
a	O
summary	O
of	O
frameworks	O
that	O
support	O
SqueezeNet	B-Application
.	O
</s>
<s>
SqueezeNet	B-Application
was	O
originally	O
described	O
in	O
a	O
paper	O
entitled	O
"	O
SqueezeNet	B-Application
:	O
AlexNet-level	O
accuracy	O
with	O
50x	O
fewer	O
parameters	O
and	O
<	O
0.5MB	O
model	O
size.	O
"	O
</s>
<s>
AlexNet	B-Algorithm
is	O
a	O
deep	O
neural	O
network	O
that	O
has	O
240	O
MB	O
of	O
parameters	O
,	O
and	O
SqueezeNet	B-Application
has	O
just	O
5	O
MB	O
of	O
parameters	O
.	O
</s>
<s>
However	O
,	O
it	O
's	O
important	O
to	O
note	O
that	O
SqueezeNet	B-Application
is	O
not	O
a	O
"	O
squeezed	O
version	O
of	O
AlexNet.	O
"	O
</s>
<s>
Rather	O
,	O
SqueezeNet	B-Application
is	O
an	O
entirely	O
different	O
DNN	O
architecture	O
than	O
AlexNet	B-Algorithm
.	O
</s>
<s>
What	O
SqueezeNet	B-Application
and	O
AlexNet	B-Algorithm
have	O
in	O
common	O
is	O
that	O
both	O
of	O
them	O
achieve	O
approximately	O
the	O
same	O
level	O
of	O
accuracy	O
when	O
evaluated	O
on	O
the	O
ImageNet	B-General_Concept
image	O
classification	O
validation	O
dataset	O
.	O
</s>
<s>
In	O
the	O
SqueezeNet	B-Application
paper	O
,	O
the	O
authors	O
demonstrated	O
that	O
a	O
model	O
compression	O
technique	O
called	O
Deep	O
Compression	O
can	O
be	O
applied	O
to	O
SqueezeNet	B-Application
to	O
further	O
reduce	O
the	O
size	O
of	O
the	O
parameter	O
file	O
from	O
5	O
MB	O
to	O
500	O
KB	O
.	O
</s>
<s>
Deep	O
Compression	O
has	O
also	O
been	O
applied	O
to	O
other	O
DNNs	O
,	O
such	O
as	O
AlexNet	B-Algorithm
and	O
VGG	O
.	O
</s>
<s>
Some	O
of	O
the	O
members	O
of	O
the	O
original	O
SqueezeNet	B-Application
team	O
have	O
continued	O
to	O
develop	O
resource-efficient	O
deep	O
neural	O
networks	O
for	O
a	O
variety	O
of	O
applications	O
.	O
</s>
<s>
As	O
with	O
the	O
original	O
SqueezeNet	B-Application
model	O
,	O
the	O
open-source	O
research	O
community	O
has	O
ported	O
and	O
adapted	O
these	O
newer	O
"	O
squeeze	O
"	O
-family	O
models	O
for	O
compatibility	O
with	O
multiple	O
deep	O
learning	O
frameworks	O
.	O
</s>
<s>
ClassificationCaffeTensorFlow	O
,	O
Keras	B-Algorithm
,	O
</s>
<s>
In	O
addition	O
,	O
the	O
open-source	O
research	O
community	O
has	O
extended	O
SqueezeNet	B-Application
to	O
other	O
applications	O
,	O
including	O
semantic	O
segmentation	O
of	O
images	O
and	O
style	B-Algorithm
transfer	I-Algorithm
.	O
</s>
