Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

.

Backward Induction
Game Theory Course:
Jackson, Leyton-Brown & Shoham

Game Theory Course: Jackson, Leyton-Brown & Shoham

Backward Induction

single depth-first traversal of the game tree, and thus requires time linear in the size
of the game representation. Recall in contrast that the best known methods for finding
Nash equilibria of general games require time exponential in the size of the normal
remember
well that the induced
form of an extensive-form
is
Idea:form;
Identify
theas equilibria
in the normal
bottom-most
trees, and game
adopt
exponentially larger than the original representation.

Computing Subgame Perfect Equilibria


these as one moves up the tree

function BACKWARD I NDUCTION (node h) returns u(h)


if h Z then
return u(h)
best util
forall a (h) do
util at child BACKWARD I NDUCTION((h, a))
if util at child(h) > best util(h) then
best util util at child
return best util

// h is a terminal node

Figure 5.6: Procedure for finding the value of a sample (subgame-perfect) Nash equi-

util_at_child
is a vector denoting
the game.
utility for each player
librium of a perfect-information
extensive-form

theThe
procedure
return an equilibrium strategy, but rather labels
algorithm Bdoesnt
ACKWARD I NDUCTION is described in Figure 5.6. The variable
each
node
with
a
vector
numbers.
util at child is a vector denotingof
thereal
utility
for each player at the child node; util at child(h)
denotes the element of this vector corresponding to the utility for player (h) (the
This
canatbe
seen
as an extension
the games
utilityfor
a vector
giving utilities
player
wholabeling
gets to move
node
h). Similarly
best util is of
eachfunction
player.
to the non-terminal nodes
that this strategies
procedure does
not areturn
equilibrium
strategy
for each of the
Observe
Equilibrium
take
bestanaction
at each
node.
players,
but rather
describes
how to label each node with a vector of n real
numbers.
Game TheorynCourse:
Jackson,
Leyton-Brown
& Shoham
Backward
Induction

single depth-first traversal of the game tree, and thus requires time linear in the size
of the game representation. Recall in contrast that the best known methods for finding
Nash equilibria of general games require time exponential in the size of the normal
remember
well that the induced
form of an extensive-form
is
Idea:form;
Identify
theas equilibria
in the normal
bottom-most
trees, and game
adopt
exponentially larger than the original representation.

Computing Subgame Perfect Equilibria


these as one moves up the tree

function BACKWARD I NDUCTION (node h) returns u(h)


if h Z then
return u(h)
best util
forall a (h) do
util at child BACKWARD I NDUCTION((h, a))
if util at child(h) > best util(h) then
best util util at child
return best util

// h is a terminal node

Figure 5.6: Procedure for finding the value of a sample (subgame-perfect) Nash equi-

of a perfect-information
extensive-form game.
Forlibrium
zero-sum
games, BackwardInduction
has another name: the
minimax algorithm.

The algorithm BACKWARD I NDUCTION is described in Figure 5.6. The variable


at child
util
vector denoting
the utility
each player
child node; util at child(h)
Here
itsis aenough
to store
onefor
number
perat the
node.
denotes
the element of this vector corresponding to the utility for player (h) (the
Its possible
to
speed
things
up
by
pruning
nodes
that will never be
player who gets to move at node h). Similarly best util is a vector giving utilities for
eachreached
player. in play: alpha-beta pruning.
Observe that this procedure does not return an equilibrium strategy for each of the
players,
but rather
describes
how to label each node with a vector of n real
numbers.
Game TheorynCourse:
Jackson,
Leyton-Brown
& Shoham
Backward
Induction

Centipede Game

118

1q

2q

(1,0)

(0,2)

Reasoning and Computing with the Extensive Form

1q

(3,1)

Figure 5.9

2q

1q

(2,4)

(4,3)

(3,5)

The centipede game

What happens when we use this procedure on Centipede?


place. In other words, you have reached a state to which your analysis has given a
In the
onlyofequilibrium,
player
goes
down
in ofthe
first
move.
probability
zero. How should you
amend1your
beliefs
and course
action
based
on
this
measure-zero
event?
It
turns
out
this
seemingly
small
inconvenience
actually
This outcome is Pareto-dominated by all but one other outcome.
raises a fundamental problem in game theory. We will not develop the subject further
here, but let us only mention that there exist different accounts of this situation, and
Two considerations:
they depend on the probabilistic assumptions made, on what is common knowledge (in
practical:
human
dont go
down right
awayhow one
particular, whether
theresubjects
is common knowledge
of rationality),
and on exactly
revises ones beliefs
in the
face of measure
zero2events.
The
last question
is intimatelygo down?
theoretical:
what
should
player
do
if
player
1
doesnt
related to the subject of belief revision discussed in Chapter 2.

5.2

SPE analysis says to go down. However, that same analysis says that
should player 2 update
beliefs
upon
observation
of
a
measure
zero
event?
Up to this point, in our discussion of extensive-form games we have allowed players to
specify
action that
wouldthat
take atplayer
every choice
nodedo
of the
game. This implies
but iftheplayer
1 they
knows
2 will
something
else, it is
that players know the node they are in, andrecalling that in such games we equate
rational
for
him
not
to
go
down
anymore...
a
paradox
nodes with the histories that led to themall the prior choices, including those of other
agents.
theres
a whole
on perfect-information
this question games.
For this
reason weliterature
have called these
P1 would already have
gone down.
How
Imperfect-information
extensive-form
games

We might not always want to make such a strong assumption about our players and
environment.
In many
with
Game Theory Course: our
Jackson,
Leyton-Brown
& situations
Shoham we may want to model agents needing to act Backward
Induction

You might also like