<s>
In	O
machine	O
learning	O
,	O
the	O
Highway	B-General_Concept
Network	I-General_Concept
was	O
the	O
first	O
working	O
very	O
deep	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
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
.	O
</s>
<s>
It	O
uses	O
skip	B-Algorithm
connections	I-Algorithm
modulated	O
by	O
learned	O
gating	O
mechanisms	O
to	O
regulate	O
information	O
flow	O
,	O
inspired	O
by	O
Long	B-Algorithm
Short-Term	I-Algorithm
Memory	I-Algorithm
(	O
LSTM	B-Algorithm
)	O
recurrent	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
.	O
</s>
<s>
The	O
advantage	O
of	O
a	O
Highway	B-General_Concept
Network	I-General_Concept
over	O
the	O
common	O
deep	O
neural	B-Architecture
networks	I-Architecture
is	O
that	O
it	O
solves	O
or	O
partially	O
prevents	O
the	O
vanishing	B-Algorithm
gradient	I-Algorithm
problem	I-Algorithm
,	O
thus	O
leading	O
to	O
easier	O
to	O
optimize	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
Highway	B-General_Concept
Networks	I-General_Concept
have	O
been	O
used	O
as	O
part	O
of	O
text	B-General_Concept
sequence	I-General_Concept
labeling	I-General_Concept
and	O
speech	B-Application
recognition	I-Application
tasks	O
.	O
</s>
<s>
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
was	O
used	O
to	O
win	O
the	O
ImageNet	O
2015	O
competition	O
.	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>
Those	O
two	O
last	O
gates	O
are	O
non-linear	O
transfer	O
functions	O
(	O
by	O
convention	O
Sigmoid	B-Algorithm
function	I-Algorithm
)	O
.	O
</s>
