STANFORD ARTIFICIAL INTELLIGENCE STAN-CS-73-391 SEARCH STRATEGIES

spineunkemptΤεχνίτη Νοημοσύνη και Ρομποτική

17 Ιουλ 2012 (πριν από 5 χρόνια και 6 μέρες)

200 εμφανίσεις

STANFORD

ARTIFICIAL

INTELLIGENCE

LABORATORY.
MEMOAlM-217
STAN-CS-73-391
SEARCH

BY
N.

SRIDHARAN
--
SUPPORTED

BY
1973
!?

NAT
CO'MPUTER

SCIENCE

DEPARTMENT
School

of

and

Sciences
STANFORD

UNIVERSITY
tNTRODUCTfON
2.
TASK ENV I RONMENT
.
3.
SOLUTION SCHEME
4.
5.
DETERMINANTS OF THE
SEARCH
SPACE
6.
SAMPLE PROBLEM AND EFFORT SPENT
DESIGN OF SEARCH STRATEGY
6a. FIXED STRATEGY
6b.
PARTIAL PATH EVALUATION
6~.

COMPLEXtTY/SIMPLICITY OF
S?JR.GOAt
COMPOUNDS
6d. SIZE OF SEARCH SPACE
6e. APPL I CATION OF KEY TRANSFORMS
6f.
SELECTION AN@ ORDERING OF ATTRIBUTES
6g. KEY I
t
,
I t i s not
I
INTRODUCTION
T h e d e s i g n
of

scientifk
L-
task such as Organic Chemical Synthesis was the
topfc
of a Doctoral
L
I
I
L
L
L
ThesFs completed in the summer of
1).
Cheml cal
synthesfs i n practi ce Fnvol ves
i)
the choice of molecule to be
synthesized;

=.
FF)
t he
formulatFon
a nd
readFly
available compounds to the target compounds with
constderatfon
of
feaslbllfty
regardi ng t he purposes of synt hesi s);
FFF)
t he sel ect i on
of
spaclftc
i ndi vi dual st eps of react i on and t hei r t emporal
executton;
iv)
t he experi ment al
execution
of the synthesfs and
adtivlty
in
iFI
above can be t ermed t he f ormal synt hesi s.
This
devel opment of the spectfi catl on of syntheses i nvol ves no l aboratory
techn!que
and is carried out mainly on paper and in the mi nds o f
chemfsts
(and now
AND
DIFFICULTY OF
CHEMlCAt

undentable
and there
scfentlsts
for synthesi s
chemtsts.
The l evel of i nt el l ect ual act i vi t y a n d
L-
WITH AN EXTENSIVE SCIENTIFIC KNOWLEDGE RASE.
Our choice is perhaps
medium-stzed

computatfon
center, and h a s
produced successfully several syntheses for each of several compl ex
molecules;
searchfng
for
solutfons
and
Incorporates--into
i t s t ask several key Judgement al
capabll
it
ies
of
a competent synthesi s
chemtst.
TASK ENVIRONMENT
i
i
The program accepts as input some representation of the target
compound t oget her
wtth
a
ltst
of condi ti ons and
constrafnts
that must
govern the proposed syntheses (Figure
1).
A list of compounds that a r e
availability)
can be consulted.
A reaction
1

ibrary
containing general
Ized
pr ocedur es
t
is suppl
ted to the program.
The output is a set of proposed syntheses,
each being a valid reaction pathway from avai l abl e compounds to the
target mol ecul e.
The syntheses are arrived at by means of
expToration
of an AND-OR search space.
The design of the search strategy
using

ANP-O?
probl em sol vi ng
c
t rees ( Ref erence
2)
concern themsel ves wi th ei ther opt i mal sol ut i ons
or mi ni mal ef f ort spent
tn f i ndi ng a sol ut i on.
exploratton
not completed by the
program are very i nformati ve as wel l.
4
Chemfsts also accept a
representatfon
o f c h e mi c a l
structures

for

program
TOPOtOGtCAL
STRUCTURE DESCRI PTI ON
far
i -
a compound.
Det ai l s of t hi s
representation
and
manfpulatfon
a r e
,
c
si t es.
The chemist knows of several reactions that can synthesize an
L
The program
Is
provided
with
a collection of react i on schemat a
called the REACTION LIBRARY.
The reaction schemata
are
grouped
Into
reacti on chapters accordi ng to the
prfarl
assi gnment of meri t rat i ng.
Bas ed on the results of ot her
t est s t he program may al t er t he meri t rat i ng t o ref l ect t he
suftability
c
of the schema to the speci fi c target mol ecul e.
L
We may represent the alternative courses of syntheses devel oped
for the target molecule by a
PROBLEM
SOLVING GRAPH (Figure
3).
The
.
target mol ecul e i s a node at the top.
A serfes of arrows lead from
the target through the chapter,attri bute and schema l ayers to the
8
/
/
COI”,lPOUND
LAYER
CHAPTER
c)
OLEFIN
BOND
43
Subgoals,

Reaction

Legend:
Filled

in
c i r c l es
Figure
5.
ONE

LEVEL

OF

SOLVING

-TED
BY

THE

compounds

found
available

in

t-
3)
are both
required for a given reaction.
All generated subgoals on the tree that
were not selected for exploration are represented by a horizontal bar,
with the number of subgoals in the unexplored group indicated under the
bar.
Subgoals that were selected for exploration that have no progeny
, on the tree (as in subgoal 8) failed to generate any subgoals that could
--.
pass the heuristic tests for admission to the search-tree.

18
Figure
6.
-.
GZNERATED

PROBLEM
SOLVING
TREE
FOR

L
Con,joined subgoal compounds A and B
A B
c
E F
c
C
)
Backup Merit for B
L
B
)
1
React i on Mer i t of E + F
1
Pr esent l y,
the functions f and
conJoined
compounds represent AND-nodes in this AND-OR tree,
.
22
and so the compound with the least merit is chosen from among
conjuncts.
Thi s i s i n accordance wi th the general strategy of
dealing with AND-OR problem solving graphs.
The eval uat ion,
backup procedure and goal selection are descr i bed
i n f ul l er det ai l s i n t he t hesi s
(
r ef er ence 1
I.
1
1
eval uat i on,
i t i s reasonabl e to choose those that are termi nated
by subgoals of higher merit.
( I f t he subgoal i s of hi gher meri t
thi s woul d i mpl y that the reacti ons are poorer on that path;
thus
one may actual l y prefer termi nati ng subgoal s wi th the l owest meri t
depending upon solution requirements.
1
SfZE

;
I t i s al so reasonabl e to use an esti mated si ze of search
L
t hat may ensue on di f f erent pat hs, i n order t o cont i nue search. I t
is especially useful when such program resources as time or storage
are dwindling or when the evaluation leaves a LARGE
NUMBFR
of
subgoal s of equal pri ori t y.
23
L
due consi derati on to
OLEFIN

BOND.
One of the operators results in a smaller but
simtlar
compound wi t h
onl y t hr ee OLEFI N
BONDS
and t he react i on i t sel f has
htgh
meri t.
When continuing search with this new subgoal a clear indication now
comes from the above observation,
to prefer to operate on another
OLEF I N
BOND.
The similarity of the
resultfng
compound al so rai ses
-=.
the expectati on that
CYEMICAL

SYNTHEWS
(suggested)
Some compounds can be changed
stmilar
L
synthesis
search may prof
I

tab1
y
L
be geared toward the
speclftc

Tntermedfate.
On the other hand,
subgoal
Is
generat ed t hat i s not avai l abl e
a
slynthesis
f or t hat i nt ermedi at e subgoal i s t o be act i vel y pursued
wi t h hi gh pr i or i t y.
L
USE OF
ANALOGY
IN CHEMICAL
SYNTFlESlS
(suggested)
Qui te often chemi sts
arrtve
at syntheses by
followTng
the known
synthesis of an analogous compound.Si tuati ons where sol uti on
(or
simpltftcation)
by anal ogy can be appl i ed
ar?se
pr of usel y:
26
the goal compound is analogous to a compound whose synthesis is
ished,
a key intermediate can be synthesized by analogy to
an avai l abl e key i nt er medi at e,a subgoal generated i s si mi l ar to one
or more intermediate compounds generated and solved by the program
duri ng thi s run al one.
However the advantages of overruling normal
search by reasoni ng through anal ogy i n these
sttuations
is not clear.
filterfng
out syntheses not desired from the output of the program.
thi s al l ows a ful l er set of
avolded
devel opi ng an i nteracti ve system
where a chemist supplies guidance on-line to the program.
Our
interest in the problem is mainly as an A? endeavour and to that
extent our
REMARKS
The strategies discussed above fal l roughl y i nto subgoal -dependence,
transform-dependence and partial-path-dependence,
The cr i t er i a t o
be used i n e a c h st rat egy ( t he l i mi t s, t hreshol ds,
orderinKs
a nd
meri t boosts) can have several sources of i nformati on
(FTgure

8).
SUBGOAL
MODEL OF PROBLEM OR
OF SOLUTION SPACE
TRANSFORM
CUMULATED PAST EXPER I ENCE
i
prescribed
to aid
i t i n f i ndi ng bet t er sol ut i ons f ast er,wi thout l eadi ng to adaptati on or
adjustment of the model.
28
Acknowledgement:
Help from Mr. Arthur Hart and Mrs. No-Jane Shue,
and
Herbert
Gelernter and Frank Fowler
IS
acknowledped
wi th deepest
than&.
I
al so thank Dee Larson for competent secretari al hel p.
‘L
L
L
29
REFERENCES
1.
Sri dharan, N.S.,
An Appl i cat i on of
Proc.

IFIP
Congress 71,
Ljubljana,
Yugosl avi a
Wipke, W.T.,
Organic Syntheses
Computer-Assisted Design of Complex
in SCIENCE, Volume 166, October 1969, p. 178-192.
30