Solving Problems with Unknown Solution Length at Almost No Extra Cost
Following up on previous work of Cathabard et al. (in: Proceedings of foundations of genetic algorithms (FOGA’11), ACM, 2011 ) we analyze variants of the (1 + 1) evolutionary algorithm (EA) for problems with unknown solution length. For their setting, in which the solution length is sampled from a g...
        Saved in:
      
    
          | Published in | Algorithmica Vol. 81; no. 2; pp. 703 - 748 | 
|---|---|
| Main Authors | , , | 
| Format | Journal Article | 
| Language | English | 
| Published | 
        New York
          Springer US
    
        15.02.2019
     Springer Nature B.V Springer Verlag  | 
| Subjects | |
| Online Access | Get full text | 
| ISSN | 0178-4617 1432-0541  | 
| DOI | 10.1007/s00453-018-0477-7 | 
Cover
| Summary: | Following up on previous work of Cathabard et al. (in: Proceedings of foundations of genetic algorithms (FOGA’11), ACM,
2011
) we analyze variants of the (1 + 1) evolutionary algorithm (EA) for problems with unknown solution length. For their setting, in which the solution length is sampled from a geometric distribution, we provide mutation rates that yield for both benchmark functions
OneMax
and
LeadingOnes
an expected optimization time that is of the same order as that of the (1 + 1) EA knowing the solution length. More than this, we show that almost the same run times can be achieved even if
no
a priori information on the solution length is available. We also regard the situation in which neither the number nor the positions of the bits with an influence on the fitness function are known. Solving an open problem from Cathabard et al. we show that, for arbitrary
s
∈
N
, such
OneMax
and
LeadingOnes
instances can be solved, simultaneously for all
n
∈
N
, in expected time
O
(
n
(
log
(
n
)
)
2
log
log
(
n
)
…
log
(
s
-
1
)
(
n
)
(
log
(
s
)
(
n
)
)
1
+
ε
)
and
O
(
n
2
log
(
n
)
log
log
(
n
)
…
log
(
s
-
1
)
(
n
)
(
log
(
s
)
(
n
)
)
1
+
ε
)
, respectively; that is, in almost the same time as if 
n
and the relevant bit positions were known. For the
LeadingOnes
case, we prove lower bounds of same asymptotic order of magnitude apart from the
(
log
(
s
)
(
n
)
)
ε
factor. Aiming at closing this arbitrarily small remaining gap, we realize that there is no asymptotically best performance for this problem. For any algorithm solving, for all 
n
, all instances of size 
n
in expected time at most
T
(
n
), there is an algorithm doing the same in time
T
′
(
n
)
with
T
′
=
o
(
T
)
. For
OneMax
we show results of similar flavor. | 
|---|---|
| Bibliography: | ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 14  | 
| ISSN: | 0178-4617 1432-0541  | 
| DOI: | 10.1007/s00453-018-0477-7 |