<s>
AlexNet	B-Algorithm
is	O
the	O
name	O
of	O
a	O
convolutional	B-Architecture
neural	I-Architecture
network	I-Architecture
(	O
CNN	B-Architecture
)	O
architecture	O
,	O
designed	O
by	O
Alex	O
Krizhevsky	O
in	O
collaboration	O
with	O
Ilya	O
Sutskever	O
and	O
Geoffrey	O
Hinton	O
,	O
who	O
was	O
Krizhevsky	O
's	O
Ph.D.	O
advisor	O
.	O
</s>
<s>
AlexNet	B-Algorithm
competed	O
in	O
the	O
ImageNet	O
Large	O
Scale	O
Visual	O
Recognition	O
Challenge	O
on	O
September	O
30	O
,	O
2012	O
.	O
</s>
<s>
The	O
original	O
paper	O
's	O
primary	O
result	O
was	O
that	O
the	O
depth	O
of	O
the	O
model	O
was	O
essential	O
for	O
its	O
high	O
performance	O
,	O
which	O
was	O
computationally	O
expensive	O
,	O
but	O
made	O
feasible	O
due	O
to	O
the	O
utilization	O
of	O
graphics	B-Architecture
processing	I-Architecture
units	I-Architecture
(	O
GPUs	B-Architecture
)	O
during	O
training	O
.	O
</s>
<s>
AlexNet	B-Algorithm
was	O
not	O
the	O
first	O
fast	O
GPU-implementation	O
of	O
a	O
CNN	B-Architecture
to	O
win	O
an	O
image	O
recognition	O
contest	O
.	O
</s>
<s>
A	O
CNN	B-Architecture
on	O
GPU	B-Architecture
by	O
K	O
.	O
Chellapilla	O
et	O
al	O
.	O
</s>
<s>
A	O
deep	O
CNN	B-Architecture
of	O
et	O
al	O
.	O
</s>
<s>
Between	O
May	O
15	O
,	O
2011	O
and	O
September	O
10	O
,	O
2012	O
,	O
their	O
CNN	B-Architecture
won	O
no	O
fewer	O
than	O
four	O
image	O
competitions	O
.	O
</s>
<s>
According	O
to	O
the	O
AlexNet	B-Algorithm
paper	O
,	O
Cireșan	O
's	O
earlier	O
net	O
is	O
"	O
somewhat	O
similar.	O
"	O
</s>
<s>
Both	O
were	O
originally	O
written	O
with	O
CUDA	B-Architecture
to	O
run	O
with	O
GPU	B-Architecture
support	O
.	O
</s>
<s>
In	O
fact	O
,	O
both	O
are	O
actually	O
just	O
variants	O
of	O
the	O
CNN	B-Architecture
designs	O
introduced	O
by	O
Yann	O
LeCun	O
et	O
al	O
.	O
</s>
<s>
(	O
1989	O
)	O
who	O
applied	O
the	O
backpropagation	B-Algorithm
algorithm	O
to	O
a	O
variant	O
of	O
Kunihiko	O
Fukushima	O
's	O
original	O
CNN	B-Architecture
architecture	O
called	O
"	O
neocognitron.	O
"	O
</s>
<s>
In	O
2015	O
,	O
AlexNet	B-Algorithm
was	O
outperformed	O
by	O
Microsoft	O
Research	O
Asia	O
's	O
very	O
deep	O
CNN	B-Architecture
with	O
over	O
100	O
layers	O
,	O
which	O
won	O
the	O
ImageNet	O
2015	O
contest	O
.	O
</s>
<s>
AlexNet	B-Algorithm
contained	O
eight	O
layers	O
;	O
the	O
first	O
five	O
were	O
convolutional	B-Language
layers	O
,	O
some	O
of	O
them	O
followed	O
by	O
max-pooling	O
layers	O
,	O
and	O
the	O
last	O
three	O
were	O
fully	O
connected	O
layers	O
.	O
</s>
<s>
The	O
network	O
,	O
except	O
the	O
last	O
layer	O
,	O
is	O
split	O
into	O
two	O
copies	O
,	O
each	O
run	O
on	O
one	O
GPU	B-Architecture
.	O
</s>
<s>
It	O
used	O
the	O
non-saturating	O
ReLU	B-Algorithm
activation	O
function	O
,	O
which	O
showed	O
improved	O
training	O
performance	O
over	O
tanh	O
and	O
sigmoid	B-Algorithm
.	O
</s>
<s>
AlexNet	B-Algorithm
is	O
considered	O
one	O
of	O
the	O
most	O
influential	O
papers	O
published	O
in	O
computer	O
vision	O
,	O
having	O
spurred	O
many	O
more	O
papers	O
published	O
employing	O
CNNs	B-Architecture
and	O
GPUs	B-Architecture
to	O
accelerate	O
deep	B-Algorithm
learning	I-Algorithm
.	O
</s>
<s>
As	O
of	O
early	O
2023	O
,	O
the	O
AlexNet	B-Algorithm
paper	O
has	O
been	O
cited	O
over	O
120,000	O
times	O
according	O
to	O
Google	O
Scholar	O
.	O
</s>
