<s>
The	O
simplest	O
kind	O
of	O
feedforward	B-Algorithm
neural	I-Algorithm
network	I-Algorithm
is	O
a	O
linear	O
network	O
,	O
which	O
consists	O
of	O
a	O
single	O
layer	O
of	O
output	O
nodes	O
;	O
the	O
inputs	O
are	O
fed	O
directly	O
to	O
the	O
outputs	O
via	O
a	O
series	O
of	O
weights	O
.	O
</s>
<s>
The	O
mean	B-Algorithm
squared	I-Algorithm
errors	I-Algorithm
between	O
these	O
calculated	O
outputs	O
and	O
a	O
given	O
target	O
values	O
are	O
minimized	O
by	O
creating	O
an	O
adjustment	O
to	O
the	O
weights	O
.	O
</s>
<s>
This	O
technique	O
has	O
been	O
known	O
for	O
over	O
two	O
centuries	O
as	O
the	O
method	B-Algorithm
of	I-Algorithm
least	I-Algorithm
squares	I-Algorithm
or	O
linear	B-General_Concept
regression	I-General_Concept
.	O
</s>
<s>
It	O
was	O
used	O
as	O
a	O
means	O
of	O
finding	O
a	O
good	O
rough	O
linear	B-General_Concept
fit	I-General_Concept
to	O
a	O
set	O
of	O
points	O
by	O
Legendre	O
(	O
1805	O
)	O
and	O
Gauss	B-Algorithm
(	O
1795	O
)	O
for	O
the	O
prediction	O
of	O
planetary	O
movement	O
.	O
</s>
<s>
Wilhelm	O
Lenz	O
and	O
Ernst	O
Ising	O
created	O
and	O
analyzed	O
the	O
Ising	O
model	O
(	O
1925	O
)	O
which	O
is	O
essentially	O
a	O
non-learning	O
artificial	O
recurrent	B-Algorithm
neural	I-Algorithm
network	I-Algorithm
(	O
RNN	O
)	O
consisting	O
of	O
neuron-like	O
threshold	O
elements	O
.	O
</s>
<s>
Warren	O
McCulloch	O
and	O
Walter	O
Pitts	O
(	O
1943	O
)	O
also	O
considered	O
a	O
non-learning	O
computational	O
model	O
for	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
One	O
approach	O
focused	O
on	O
biological	O
processes	O
while	O
the	O
other	O
focused	O
on	O
the	O
application	O
of	O
neural	B-Architecture
networks	I-Architecture
to	O
artificial	B-Application
intelligence	I-Application
.	I-Application
</s>
<s>
This	O
work	O
led	O
to	O
work	O
on	O
nerve	O
networks	O
and	O
their	O
link	O
to	O
finite	B-Architecture
automata	I-Architecture
.	O
</s>
<s>
Hebbian	O
learning	O
is	O
unsupervised	B-General_Concept
learning	I-General_Concept
.	O
</s>
<s>
Other	O
neural	B-Architecture
network	I-Architecture
computational	O
machines	O
were	O
created	O
by	O
Rochester	O
,	O
Holland	O
,	O
Habit	O
and	O
Duda	O
(	O
1956	O
)	O
.	O
</s>
<s>
Rosenblatt	O
(	O
1958	O
)	O
created	O
the	O
perceptron	B-Algorithm
,	O
an	O
algorithm	O
for	O
pattern	O
recognition	O
.	O
</s>
<s>
With	O
mathematical	O
notation	O
,	O
Rosenblatt	O
described	O
circuitry	O
not	O
in	O
the	O
basic	O
perceptron	B-Algorithm
,	O
such	O
as	O
the	O
exclusive-or	O
circuit	O
that	O
could	O
not	O
be	O
processed	O
by	O
neural	B-Architecture
networks	I-Architecture
at	O
the	O
time	O
.	O
</s>
<s>
Some	O
say	O
that	O
research	O
stagnated	O
following	O
Minsky	O
and	O
Papert	O
(	O
1969	O
)	O
,	O
who	O
discovered	O
that	O
basic	O
perceptrons	B-Algorithm
were	O
incapable	O
of	O
processing	O
the	O
exclusive-or	O
circuit	O
and	O
that	O
computers	O
lacked	O
sufficient	O
power	O
to	O
process	O
useful	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
However	O
,	O
by	O
the	O
time	O
this	O
book	O
came	O
out	O
,	O
methods	O
for	O
training	O
multilayer	B-Algorithm
perceptrons	I-Algorithm
(	O
MLPs	O
)	O
by	O
deep	B-Algorithm
learning	I-Algorithm
were	O
already	O
known	O
.	O
</s>
<s>
The	O
first	O
deep	B-Algorithm
learning	I-Algorithm
MLP	O
was	O
published	O
by	O
Alexey	O
Grigorevich	O
Ivakhnenko	O
and	O
Valentin	O
Lapa	O
in	O
1965	O
,	O
as	O
the	O
Group	B-Algorithm
Method	I-Algorithm
of	I-Algorithm
Data	I-Algorithm
Handling	I-Algorithm
.	O
</s>
<s>
The	O
first	O
deep	B-Algorithm
learning	I-Algorithm
MLP	O
trained	O
by	O
stochastic	B-Algorithm
gradient	I-Algorithm
descent	I-Algorithm
was	O
published	O
in	O
1967	O
by	O
Shun'ichi	O
Amari	O
.	O
</s>
<s>
The	O
backpropagation	B-Algorithm
algorithm	O
is	O
an	O
efficient	O
application	O
of	O
the	O
Leibniz	O
chain	O
rule	O
(	O
1673	O
)	O
to	O
networks	O
of	O
differentiable	O
nodes	O
.	O
</s>
<s>
the	O
reverse	O
mode	O
of	O
automatic	B-Algorithm
differentiation	I-Algorithm
or	O
reverse	O
accumulation	O
,	O
due	O
to	O
Seppo	O
Linnainmaa	O
(	O
1970	O
)	O
.	O
</s>
<s>
The	O
term	O
"	O
back-propagating	O
errors	O
"	O
was	O
introduced	O
in	O
1962	O
by	O
Frank	O
Rosenblatt	O
,	O
but	O
he	O
did	O
not	O
have	O
an	O
implementation	O
of	O
this	O
procedure	O
,	O
although	O
Henry	O
J	O
.	O
Kelley	O
had	O
a	O
continuous	O
precursor	O
of	O
backpropagation	B-Algorithm
already	O
in	O
1960	O
in	O
the	O
context	O
of	O
control	O
theory	O
.	O
</s>
<s>
In	O
1982	O
,	O
Paul	O
Werbos	O
applied	O
backpropagation	B-Algorithm
to	O
MLPs	O
in	O
the	O
way	O
that	O
has	O
become	O
standard	O
.	O
</s>
<s>
Self-organizing	B-Algorithm
maps	I-Algorithm
(	O
SOMs	O
)	O
were	O
described	O
by	O
Teuvo	O
Kohonen	B-Algorithm
in	O
1982	O
.	O
</s>
<s>
SOMs	O
are	O
neurophysiologically	O
inspired	O
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
that	O
learn	O
low-dimensional	B-Algorithm
representations	O
of	O
high-dimensional	O
data	O
while	O
preserving	O
the	O
topological	B-Architecture
structure	I-Architecture
of	O
the	O
data	O
.	O
</s>
<s>
They	O
are	O
trained	O
using	O
competitive	B-Algorithm
learning	I-Algorithm
.	O
</s>
<s>
Support	B-Algorithm
vector	I-Algorithm
machines	I-Algorithm
,	O
developed	O
at	O
AT&T	O
Bell	O
Laboratories	O
by	O
Vladimir	O
Vapnik	O
with	O
colleagues	O
(	O
Boser	O
et	O
al.	O
,	O
1992	O
,	O
Isabelle	O
Guyon	O
et	O
al.	O
,	O
1993	O
,	O
Corinna	O
Cortes	O
,	O
1995	O
,	O
Vapnik	O
et	O
al.	O
,	O
1997	O
)	O
and	O
simpler	O
methods	O
such	O
as	O
linear	B-General_Concept
classifiers	I-General_Concept
gradually	O
overtook	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
However	O
,	O
neural	B-Architecture
networks	I-Architecture
transformed	O
domains	O
such	O
as	O
the	O
prediction	O
of	O
protein	O
structures	O
.	O
</s>
<s>
The	O
origin	O
of	O
the	O
CNN	O
architecture	O
is	O
the	O
"	O
neocognitron	B-Algorithm
"	O
introduced	O
by	O
Kunihiko	O
Fukushima	O
in	O
1980	O
.	O
</s>
<s>
The	O
neocognitron	B-Algorithm
introduced	O
the	O
two	O
basic	O
types	O
of	O
layers	O
in	O
CNNs	O
:	O
convolutional	O
layers	O
,	O
and	O
downsampling	O
layers	O
.	O
</s>
<s>
In	O
1969	O
,	O
Kunihiko	O
Fukushima	O
also	O
introduced	O
the	O
ReLU	B-Algorithm
(	O
rectified	B-Algorithm
linear	I-Algorithm
unit	I-Algorithm
)	O
activation	B-Algorithm
function	I-Algorithm
.	O
</s>
<s>
The	O
rectifier	B-Algorithm
has	O
become	O
the	O
most	O
popular	O
activation	B-Algorithm
function	I-Algorithm
for	O
CNNs	O
and	O
deep	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
in	O
general	O
.	O
</s>
<s>
The	O
time	B-Algorithm
delay	I-Algorithm
neural	I-Algorithm
network	I-Algorithm
(	O
TDNN	B-Algorithm
)	O
was	O
introduced	O
in	O
1987	O
by	O
Alex	O
Waibel	O
and	O
was	O
one	O
of	O
the	O
first	O
CNNs	O
,	O
as	O
it	O
achieved	O
shift	O
invariance	O
.	O
</s>
<s>
It	O
did	O
so	O
by	O
utilizing	O
weight	O
sharing	O
in	O
combination	O
with	O
backpropagation	B-Algorithm
training	O
.	O
</s>
<s>
Thus	O
,	O
while	O
also	O
using	O
a	O
pyramidal	O
structure	O
as	O
in	O
the	O
neocognitron	B-Algorithm
,	O
it	O
performed	O
a	O
global	O
optimization	O
of	O
the	O
weights	O
instead	O
of	O
a	O
local	O
one	O
.	O
</s>
<s>
to	O
a	O
CNN	O
(	O
a	O
simplified	O
Neocognitron	B-Algorithm
with	O
convolutional	O
interconnections	O
between	O
the	O
image	O
feature	O
layers	O
and	O
the	O
last	O
fully	O
connected	O
layer	O
)	O
for	O
alphabet	O
recognition	O
.	O
</s>
<s>
trained	O
a	O
CNN	O
with	O
the	O
purpose	O
of	O
recognizing	O
handwritten	O
ZIP	B-Language
codes	O
on	O
mail	O
.	O
</s>
<s>
They	O
combined	O
TDNNs	B-Algorithm
with	O
max-pooling	O
in	O
order	O
to	O
realize	O
a	O
speaker	O
independent	O
isolated	O
word	O
recognition	O
system	O
.	O
</s>
<s>
In	O
a	O
variant	O
of	O
the	O
neocognitron	B-Algorithm
called	O
the	O
cresceptron	O
,	O
instead	O
of	O
using	O
Fukushima	O
's	O
spatial	O
averaging	O
,	O
J	O
.	O
Weng	O
et	O
al	O
.	O
</s>
<s>
In	O
2010	O
,	O
Backpropagation	B-Algorithm
training	O
through	O
max-pooling	O
was	O
accelerated	O
by	O
GPUs	O
and	O
shown	O
to	O
perform	O
better	O
than	O
other	O
pooling	O
variants	O
.	O
</s>
<s>
Behnke	O
(	O
2003	O
)	O
relied	O
only	O
on	O
the	O
sign	O
of	O
the	O
gradient	O
(	O
Rprop	B-Algorithm
)	O
on	O
problems	O
such	O
as	O
image	O
reconstruction	O
and	O
face	O
localization	O
.	O
</s>
<s>
Rprop	B-Algorithm
is	O
a	O
first-order	B-Algorithm
optimization	O
algorithm	O
created	O
by	O
Martin	O
Riedmiller	O
and	O
Heinrich	O
Braun	O
in	O
1992	O
.	O
</s>
<s>
In	O
2011	O
,	O
a	O
deep	O
GPU-based	O
CNN	O
called	O
"	O
DanNet	O
"	O
by	O
Dan	O
Ciresan	O
,	O
Ueli	O
Meier	O
,	O
and	O
Juergen	O
Schmidhuber	O
achieved	O
human-competitive	O
performance	O
for	O
the	O
first	O
time	O
in	O
computer	B-Application
vision	I-Application
contests	O
.	O
</s>
<s>
In	O
1991	O
,	O
Juergen	O
Schmidhuber	O
published	O
adversarial	O
neural	B-Architecture
networks	I-Architecture
that	O
contest	O
with	O
each	O
other	O
in	O
the	O
form	O
of	O
a	O
zero-sum	O
game	O
,	O
where	O
one	O
network	O
's	O
gain	O
is	O
the	O
other	O
network	O
's	O
loss	O
.	O
</s>
<s>
The	O
second	O
network	O
learns	O
by	O
gradient	B-Algorithm
descent	I-Algorithm
to	O
predict	O
the	O
reactions	O
of	O
the	O
environment	O
to	O
these	O
patterns	O
.	O
</s>
<s>
Earlier	O
adversarial	O
machine	O
learning	O
systems	O
"	O
neither	O
involved	O
unsupervised	O
neural	B-Architecture
networks	I-Architecture
nor	O
were	O
about	O
modeling	O
data	O
nor	O
used	O
gradient	O
descent.	O
"	O
</s>
<s>
In	O
2014	O
,	O
this	O
adversarial	O
principle	O
was	O
used	O
in	O
a	O
generative	B-Algorithm
adversarial	I-Algorithm
network	I-Algorithm
(	O
GAN	O
)	O
by	O
Ian	O
Goodfellow	O
et	O
al	O
.	O
</s>
<s>
This	O
can	O
be	O
used	O
to	O
create	O
realistic	O
deepfakes	B-Application
.	O
</s>
<s>
In	O
1992	O
,	O
Schmidhuber	O
also	O
published	O
another	O
type	O
of	O
gradient-based	O
adversarial	O
neural	B-Architecture
networks	I-Architecture
where	O
the	O
goal	O
of	O
the	O
zero-sum	O
game	O
is	O
to	O
create	O
disentangled	O
representations	O
of	O
input	O
patterns	O
.	O
</s>
<s>
Nvidia	O
's	O
StyleGAN	B-Application
(	O
2018	O
)	O
is	O
based	O
on	O
the	O
Progressive	O
GAN	O
by	O
Tero	O
Karras	O
,	O
Timo	O
Aila	O
,	O
Samuli	O
Laine	O
,	O
and	O
Jaakko	O
Lehtinen	O
.	O
</s>
<s>
StyleGANs	B-Application
improve	O
consistency	O
between	O
fine	O
and	O
coarse	O
details	O
in	O
the	O
generator	O
network	O
.	O
</s>
<s>
Many	O
modern	O
large	O
language	O
models	O
such	O
as	O
ChatGPT	B-General_Concept
,	O
GPT-4	O
,	O
and	O
BERT	B-General_Concept
use	O
a	O
feedforward	B-Algorithm
neural	I-Algorithm
network	I-Algorithm
called	O
Transformer	B-Algorithm
by	O
Ashish	O
Vaswani	O
et	O
.	O
</s>
<s>
Transformers	B-Algorithm
have	O
increasingly	O
become	O
the	O
model	O
of	O
choice	O
for	O
natural	B-Language
language	I-Language
processing	I-Language
problems	O
,	O
replacing	O
recurrent	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
(	O
RNNs	O
)	O
such	O
as	O
long	B-Algorithm
short-term	I-Algorithm
memory	I-Algorithm
(	O
LSTM	B-Algorithm
)	O
.	O
</s>
<s>
Basic	O
ideas	O
for	O
this	O
go	O
back	O
a	O
long	O
way	O
:	O
in	O
1992	O
,	O
Juergen	O
Schmidhuber	O
published	O
the	O
Transformer	B-Algorithm
with	O
"	O
linearized	O
self-attention	O
"	O
(	O
save	O
for	O
a	O
normalization	O
operator	O
)	O
,	O
</s>
<s>
which	O
is	O
also	O
called	O
the	O
"	O
linear	O
Transformer.	O
"	O
</s>
<s>
Here	O
a	O
slow	O
feedforward	B-Algorithm
neural	I-Algorithm
network	I-Algorithm
learns	O
by	O
gradient	B-Algorithm
descent	I-Algorithm
to	O
control	O
the	O
fast	O
weights	O
of	O
another	O
neural	B-Architecture
network	I-Architecture
through	O
outer	O
products	O
of	O
self-generated	O
activation	O
patterns	O
called	O
"	O
FROM	O
"	O
and	O
"	O
TO	O
"	O
which	O
in	O
Transformer	B-Algorithm
terminology	O
are	O
called	O
"	O
key	O
"	O
and	O
"	O
value	O
"	O
for	O
"	O
self-attention.	O
"	O
</s>
<s>
The	O
2017	O
Transformer	B-Algorithm
combines	O
this	O
with	O
a	O
softmax	O
operator	O
and	O
a	O
projection	O
matrix	O
.	O
</s>
<s>
Transformers	B-Algorithm
are	O
also	O
increasingly	O
being	O
used	O
in	O
computer	B-Application
vision	I-Application
.	O
</s>
<s>
In	O
the	O
1980s	O
,	O
backpropagation	B-Algorithm
did	O
not	O
work	O
well	O
for	O
deep	O
FNNs	O
and	O
RNNs	O
.	O
</s>
<s>
More	O
precisely	O
,	O
deep	B-Algorithm
learning	I-Algorithm
systems	O
have	O
a	O
substantial	O
credit	O
assignment	O
path	O
(	O
CAP	O
)	O
depth	O
.	O
</s>
<s>
To	O
overcome	O
this	O
problem	O
,	O
Juergen	O
Schmidhuber	O
(	O
1992	O
)	O
proposed	O
a	O
self-supervised	O
hierarchy	O
of	O
RNNs	O
pre-trained	O
one	O
level	O
at	O
a	O
time	O
by	O
self-supervised	B-General_Concept
learning	I-General_Concept
.	O
</s>
<s>
In	O
1993	O
,	O
a	O
chunker	O
solved	O
a	O
deep	B-Algorithm
learning	I-Algorithm
task	O
whose	O
CAP	O
depth	O
exceeded	O
1000	O
.	O
</s>
<s>
Such	O
history	O
compressors	O
can	O
substantially	O
facilitate	O
downstream	O
supervised	O
deep	B-Algorithm
learning	I-Algorithm
.	O
</s>
<s>
(	O
2006	O
)	O
proposed	O
learning	O
a	O
high-level	O
internal	O
representation	O
using	O
successive	O
layers	O
of	O
binary	O
or	O
real-valued	O
latent	O
variables	O
with	O
a	O
restricted	B-Algorithm
Boltzmann	I-Algorithm
machine	I-Algorithm
to	O
model	O
each	O
layer	O
.	O
</s>
<s>
This	O
RBM	O
is	O
a	O
generative	O
stochastic	O
feedforward	B-Algorithm
neural	I-Algorithm
network	I-Algorithm
that	O
can	O
learn	O
a	O
probability	O
distribution	O
over	O
its	O
set	O
of	O
inputs	O
.	O
</s>
<s>
In	O
2012	O
,	O
Andrew	O
Ng	O
and	O
Jeff	O
Dean	O
created	O
an	O
FNN	O
that	O
learned	O
to	O
recognize	O
higher-level	O
concepts	O
,	O
such	O
as	O
cats	O
,	O
only	O
from	O
watching	O
unlabeled	O
images	O
taken	O
from	O
YouTube	B-General_Concept
videos	I-General_Concept
.	O
</s>
<s>
Hochreiter	O
not	O
only	O
tested	O
the	O
neural	O
history	O
compressor	O
,	O
but	O
also	O
identified	O
and	O
analyzed	O
the	O
vanishing	B-Algorithm
gradient	I-Algorithm
problem	I-Algorithm
.	O
</s>
<s>
He	O
proposed	O
recurrent	O
residual	B-Algorithm
connections	O
to	O
solve	O
this	O
problem	O
.	O
</s>
<s>
This	O
led	O
to	O
the	O
deep	B-Algorithm
learning	I-Algorithm
method	O
called	O
long	B-Algorithm
short-term	I-Algorithm
memory	I-Algorithm
(	O
LSTM	B-Algorithm
)	O
,	O
published	O
in	O
1997	O
.	O
</s>
<s>
LSTM	B-Algorithm
recurrent	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
can	O
learn	O
"	O
very	O
deep	B-Algorithm
learning	I-Algorithm
"	O
tasks	O
with	O
long	O
credit	O
assignment	O
paths	O
that	O
require	O
memories	O
of	O
events	O
that	O
happened	O
thousands	O
of	O
discrete	O
time	O
steps	O
before	O
.	O
</s>
<s>
The	O
"	O
vanilla	O
LSTM	B-Algorithm
"	O
with	O
forget	O
gate	O
was	O
introduced	O
in	O
1999	O
by	O
Felix	O
Gers	O
,	O
Schmidhuber	O
and	O
Fred	O
Cummins	O
.	O
</s>
<s>
LSTM	B-Algorithm
has	O
become	O
the	O
most	O
cited	O
neural	B-Architecture
network	I-Architecture
of	O
the	O
20th	O
century	O
.	O
</s>
<s>
In	O
2015	O
,	O
Rupesh	O
Kumar	O
Srivastava	O
,	O
Klaus	O
Greff	O
,	O
and	O
Schmidhuber	O
used	O
LSTM	B-Algorithm
principles	O
to	O
create	O
the	O
Highway	B-General_Concept
network	I-General_Concept
,	O
a	O
feedforward	B-Algorithm
neural	I-Algorithm
network	I-Algorithm
with	O
hundreds	O
of	O
layers	O
,	O
much	O
deeper	O
than	O
previous	O
networks	O
.	O
</s>
<s>
7	O
months	O
later	O
,	O
Kaiming	O
He	O
,	O
Xiangyu	O
Zhang	O
;	O
Shaoqing	O
Ren	O
,	O
and	O
Jian	O
Sun	O
won	O
the	O
ImageNet	O
2015	O
competition	O
with	O
an	O
open-gated	O
or	O
gateless	O
Highway	B-General_Concept
network	I-General_Concept
variant	O
called	O
Residual	B-Algorithm
neural	I-Algorithm
network	I-Algorithm
.	O
</s>
<s>
This	O
has	O
become	O
the	O
most	O
cited	O
neural	B-Architecture
network	I-Architecture
of	O
the	O
21st	O
century	O
.	O
</s>
<s>
In	O
2011	O
,	O
Xavier	O
Glorot	O
,	O
Antoine	O
Bordes	O
and	O
Yoshua	O
Bengio	O
found	O
that	O
the	O
ReLU	B-Algorithm
of	O
Kunihiko	O
Fukushima	O
also	O
helps	O
to	O
overcome	O
the	O
vanishing	B-Algorithm
gradient	I-Algorithm
problem	I-Algorithm
,	O
compared	O
to	O
widely	O
used	O
activation	B-Algorithm
functions	I-Algorithm
prior	O
to	O
2011	O
.	O
</s>
<s>
The	O
development	O
of	O
metal	B-Architecture
–	I-Architecture
oxide	I-Architecture
–	I-Architecture
semiconductor	I-Architecture
(	O
MOS	O
)	O
very-large-scale	O
integration	O
(	O
VLSI	O
)	O
,	O
combining	O
millions	O
or	O
billions	O
of	O
MOS	B-Architecture
transistors	I-Architecture
onto	O
a	O
single	O
chip	O
in	O
the	O
form	O
of	O
complementary	O
MOS	O
(	O
CMOS	B-Device
)	O
technology	O
,	O
enabled	O
the	O
development	O
of	O
practical	O
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
in	O
the	O
1980s	O
.	O
</s>
<s>
Computational	O
devices	O
were	O
created	O
in	O
CMOS	B-Device
,	O
for	O
both	O
biophysical	O
simulation	O
and	O
neuromorphic	O
computing	O
inspired	O
by	O
the	O
structure	O
and	O
function	O
of	O
the	O
human	O
brain	O
.	O
</s>
<s>
Nanodevices	O
for	O
very	O
large	O
scale	O
principal	B-Application
components	I-Application
analyses	O
and	O
convolution	B-Language
may	O
create	O
a	O
new	O
class	O
of	O
neural	B-Architecture
computing	I-Architecture
because	O
they	O
are	O
fundamentally	O
analog	O
rather	O
than	O
digital	B-General_Concept
(	O
even	O
though	O
the	O
first	O
implementations	O
may	O
use	O
digital	B-General_Concept
devices	O
)	O
.	O
</s>
<s>
Ciresan	O
and	O
colleagues	O
(	O
2010	O
)	O
in	O
Schmidhuber	O
's	O
group	O
showed	O
that	O
despite	O
the	O
vanishing	B-Algorithm
gradient	I-Algorithm
problem	I-Algorithm
,	O
GPUs	O
make	O
backpropagation	B-Algorithm
feasible	O
for	O
many-layered	O
feedforward	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
.	O
</s>
<s>
Between	O
2009	O
and	O
2012	O
,	O
recurrent	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
and	O
deep	O
feedforward	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
developed	O
in	O
Schmidhuber	O
's	O
research	O
group	O
won	O
eight	O
international	O
competitions	O
in	O
pattern	O
recognition	O
and	O
machine	O
learning	O
.	O
</s>
<s>
For	O
example	O
,	O
the	O
bi-directional	O
and	O
multi-dimensional	O
long	B-Algorithm
short-term	I-Algorithm
memory	I-Algorithm
(	O
LSTM	B-Algorithm
)	O
of	O
Graves	O
et	O
al	O
.	O
</s>
<s>
won	O
three	O
competitions	O
in	O
connected	O
handwriting	B-Application
recognition	I-Application
at	O
the	O
2009	O
International	O
Conference	O
on	O
Document	O
Analysis	O
and	O
Recognition	O
(	O
ICDAR	O
)	O
,	O
without	O
any	O
prior	O
knowledge	O
about	O
the	O
three	O
languages	O
to	O
be	O
learned	O
.	O
</s>
<s>
Their	O
neural	B-Architecture
networks	I-Architecture
were	O
the	O
first	O
pattern	O
recognizers	O
to	O
achieve	O
human-competitive/superhuman	O
performance	O
on	O
benchmarks	O
such	O
as	O
traffic	O
sign	O
recognition	O
(	O
IJCNN	O
2012	O
)	O
,	O
or	O
the	O
MNIST	B-General_Concept
handwritten	I-General_Concept
digits	I-General_Concept
problem	I-General_Concept
.	O
</s>
<s>
Researchers	O
demonstrated	O
(	O
2010	O
)	O
that	O
deep	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
interfaced	O
to	O
a	O
hidden	O
Markov	O
model	O
with	O
context-dependent	O
states	O
that	O
define	O
the	O
neural	B-Architecture
network	I-Architecture
output	O
layer	O
can	O
drastically	O
reduce	O
errors	O
in	O
large-vocabulary	O
speech	O
recognition	O
tasks	O
such	O
as	O
voice	O
search	O
.	O
</s>
<s>
Deep	O
,	O
highly	O
nonlinear	O
neural	O
architectures	O
similar	O
to	O
the	O
neocognitron	B-Algorithm
and	O
the	O
"	O
standard	O
architecture	O
of	O
vision	O
"	O
,	O
inspired	O
by	O
simple	O
and	O
complex	O
cells	O
,	O
were	O
pre-trained	O
with	O
unsupervised	O
methods	O
by	O
Hinton	O
.	O
</s>
