<s>
Energy-based	B-General_Concept
generative	I-General_Concept
neural	I-General_Concept
networks	I-General_Concept
is	O
a	O
class	O
of	O
generative	O
models	O
,	O
which	O
aim	O
to	O
learn	O
explicit	O
probability	O
distributions	O
of	O
data	O
in	O
the	O
form	O
of	O
energy-based	O
models	O
whose	O
energy	O
functions	O
are	O
parameterized	O
by	O
modern	O
deep	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
Its	O
name	O
is	O
due	O
to	O
the	O
fact	O
that	O
this	O
model	O
can	O
be	O
derived	O
from	O
the	O
discriminative	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
The	O
parameter	O
of	O
the	O
neural	B-Architecture
network	I-Architecture
in	O
this	O
model	O
is	O
trained	O
in	O
a	O
generative	O
manner	O
by	O
Markov	O
chain	O
Monte	O
Carlo(MCMC )	O
-based	O
maximum	O
likelihood	O
estimation	O
.	O
</s>
<s>
The	O
learning	O
process	O
follows	O
an	O
''	O
analysis	O
by	O
synthesis''	O
scheme	O
,	O
where	O
within	O
each	O
learning	O
iteration	O
,	O
the	O
algorithm	O
samples	O
the	O
synthesized	O
examples	O
from	O
the	O
current	O
model	O
by	O
a	O
gradient-based	O
MCMC	B-General_Concept
method	I-General_Concept
,	O
e.g.	O
,	O
Langevin	O
dynamics	O
,	O
and	O
then	O
updates	O
the	O
model	O
parameters	O
based	O
on	O
the	O
difference	O
between	O
the	O
training	O
examples	O
and	O
the	O
synthesized	O
ones	O
.	O
</s>
<s>
The	O
first	O
energy-based	B-General_Concept
generative	I-General_Concept
neural	I-General_Concept
network	I-General_Concept
is	O
the	O
generative	O
ConvNet	B-Architecture
proposed	O
in	O
2016	O
for	O
image	O
patterns	O
,	O
where	O
the	O
neural	B-Architecture
network	I-Architecture
is	O
a	O
convolutional	B-Architecture
neural	I-Architecture
network	I-Architecture
.	O
</s>
