<s>
Action	B-Algorithm
model	I-Algorithm
learning	I-Algorithm
(	O
sometimes	O
abbreviated	O
action	O
learning	O
)	O
is	O
an	O
area	O
of	O
machine	O
learning	O
concerned	O
with	O
creation	O
and	O
modification	O
of	O
software	B-General_Concept
agent	I-General_Concept
's	O
knowledge	O
about	O
effects	O
and	O
preconditions	O
of	O
the	O
actions	O
that	O
can	O
be	O
executed	O
within	O
its	O
environment	O
.	O
</s>
<s>
This	O
knowledge	O
is	O
usually	O
represented	O
in	O
logic-based	O
action	B-Language
description	I-Language
language	I-Language
and	O
used	O
as	O
the	O
input	O
for	O
automated	B-Application
planners	I-Application
.	O
</s>
<s>
Action	B-Algorithm
model	I-Algorithm
learning	I-Algorithm
is	O
a	O
form	O
of	O
inductive	O
reasoning	O
,	O
where	O
new	O
knowledge	O
is	O
generated	O
based	O
on	O
agent	O
's	O
observations	O
.	O
</s>
<s>
It	O
differs	O
from	O
standard	O
supervised	B-General_Concept
learning	I-General_Concept
in	O
that	O
correct	O
input/output	O
pairs	O
are	O
never	O
presented	O
,	O
nor	O
imprecise	O
action	O
models	O
explicitly	O
corrected	O
.	O
</s>
<s>
Usual	O
motivation	O
for	O
action	B-Algorithm
model	I-Algorithm
learning	I-Algorithm
is	O
the	O
fact	O
that	O
manual	O
specification	O
of	O
action	O
models	O
for	O
planners	O
is	O
often	O
a	O
difficult	O
,	O
time	O
consuming	O
,	O
and	O
error-prone	O
task	O
(	O
especially	O
in	O
complex	O
environments	O
)	O
.	O
</s>
<s>
Given	O
a	O
training	O
set	O
consisting	O
of	O
examples	O
,	O
where	O
are	O
observations	O
of	O
a	O
world	O
state	O
from	O
two	O
consecutive	O
time	O
steps	O
and	O
is	O
an	O
action	O
instance	O
observed	O
in	O
time	O
step	O
,	O
the	O
goal	O
of	O
action	B-Algorithm
model	I-Algorithm
learning	I-Algorithm
in	O
general	O
is	O
to	O
construct	O
an	O
action	O
model	O
,	O
where	O
is	O
a	O
description	O
of	O
domain	O
dynamics	O
in	O
action	O
description	O
formalism	O
like	O
STRIPS	B-Application
,	O
ADL	B-Architecture
or	O
PDDL	B-Application
and	O
is	O
a	O
probability	O
function	O
defined	O
over	O
the	O
elements	O
of	O
.	O
</s>
<s>
Recent	O
action	O
learning	O
methods	O
take	O
various	O
approaches	O
and	O
employ	O
a	O
wide	O
variety	O
of	O
tools	O
from	O
different	O
areas	O
of	O
artificial	B-Application
intelligence	I-Application
and	O
computational	O
logic	O
.	O
</s>
<s>
Another	O
technique	O
,	O
in	O
which	O
learning	O
is	O
converted	O
into	O
a	O
satisfiability	O
problem	O
(	O
weighted	O
MAX-SAT	B-Application
in	O
this	O
case	O
)	O
and	O
SAT	B-Application
solvers	I-Application
are	O
used	O
,	O
is	O
implemented	O
in	O
ARMS	O
(	O
Action-Relation	O
Modeling	O
System	O
)	O
.	O
</s>
<s>
Two	O
mutually	O
similar	O
,	O
fully	O
declarative	O
approaches	O
to	O
action	O
learning	O
were	O
based	O
on	O
logic	O
programming	O
paradigm	O
Answer	B-Application
Set	I-Application
Programming	I-Application
(	O
ASP	O
)	O
and	O
its	O
extension	O
,	O
Reactive	O
ASP	O
.	O
</s>
<s>
In	O
another	O
example	O
,	O
bottom-up	O
inductive	B-Application
logic	I-Application
programming	I-Application
approach	O
was	O
employed	O
.	O
</s>
<s>
In	O
the	O
older	O
paper	O
from	O
1992	O
,	O
the	O
action	B-Algorithm
model	I-Algorithm
learning	I-Algorithm
was	O
studied	O
as	O
an	O
extension	O
of	O
reinforcement	O
learning	O
.	O
</s>
<s>
Most	O
action	O
learning	O
research	O
papers	O
are	O
published	O
in	O
journals	O
and	O
conferences	O
focused	O
on	O
artificial	B-Application
intelligence	I-Application
in	O
general	O
(	O
e.g.	O
</s>
<s>
Journal	O
of	O
Artificial	B-Application
Intelligence	I-Application
Research	I-Application
(	O
JAIR	O
)	O
,	O
Artificial	B-Application
Intelligence	I-Application
,	O
Applied	O
Artificial	B-Application
Intelligence	I-Application
(	O
AAI	O
)	O
or	O
AAAI	O
conferences	O
)	O
.	O
</s>
<s>
Despite	O
mutual	O
relevance	O
of	O
the	O
topics	O
,	O
action	B-Algorithm
model	I-Algorithm
learning	I-Algorithm
is	O
usually	O
not	O
addressed	O
on	O
planning	B-Application
conferences	O
like	O
ICAPS	O
.	O
</s>
