Bandit Algorithms

(Jeff_L) #1
14.3 Notes 182

This indeed is sufficient since



p∧q=


Ap∧q+


Acp∧q≤


Ap+


Acq=
P(A) +Q(Ac). We start with an inequality attributed to French mathematician
Lucien Le Cam, which lower bounds the left-hand side of Eq. (14.7). The inequality
states that

p∧q≥

1


2


(∫



pq

) 2


. (14.8)


Starting from the right-hand side above usingpq= (p∧q)(p∨q) and Cauchy-
Schwarz we get
(∫

pq

) 2


=


(∫ √


(p∧q)(p∨q)

) 2



(∫


p∧q

) (∫


p∨q

)


Now, using∫ p∧q+p∨q = p+q, the proof is finished by substituting
p∨q= 2−


p∧q≤2 and dividing both sides by two. It remains to lower
bound the right-hand side of(14.8). For this, we use Jensen’s inequality. First,
we write (·)^2 as exp(2 log(·)) and then move the log inside the integral:
(∫

pq


) 2


= exp

(


2 log

∫ √


pq

)


= exp

(


2 log


p

√q

p

)


≥exp

(


2



p^1
2

log

(


q
p

))


= exp

(




pq> 0

plog

(


p
q

))


= exp

(




plog

(


p
q

))


= exp (−D(P,Q)).

In the fourth and the last step we used that sincePQ,q= 0 impliesp= 0
and sop >0, which impliesq >0, and eventuallypq >0. The result is completed
by chaining the inequalities.

14.3 Notes


1 A codec:N+→ { 0 , 1 }∗is uniquely decodable ifi 1 ,...,in7→c(i 1 )···c(in)
is injective, where on the right-hand side the codes are simply concatenated.
Kraft’s inequalitystates that for any uniquely decodable codec,
∑∞

i=1

2 −`(c(i))≤ 1. (14.9)

Furthermore, for any (`n)∞n=1satisfying

∑∞


i=1^2

−`≤1 there exists a prefix code
c:N+→{ 0 , 1 }∗such that`(c(i)) =`i. The second part justifies our restriction
to prefix codes rather than uniquely decodable codes in the definition of the
entropy.
2 The supremum in the definition given in Eq. (14.5) may often be taken over
a smaller set. Precisely, let (X,G) be a measurable space and suppose that
G=σ(F) whereFis a field. Note that a field is defined by the same axioms
Free download pdf