Title: Unconditional Weak derandomization of weak algorithms Explicit versions of Yao
1UnconditionalWeak derandomization of weak
algorithmsExplicit versions of Yaos lemma
- Ronen Shaltiel, University of Haifa
2Derandomization The goal
- Main open problem Show that BPPP.
(There is evidence that this is
hard IKW,KI) - More generally
- Convert randomized algorithm A(x,r)
- into deterministic algorithm B(x)
- Wed like to
- Preserve complexity
complexity(B) complexity(A) (known
BPP?EXP). - Preserve uniformity
transformation A ? B is explicit (known
BPP?P/poly).
n bit long input
m bit long coin tosses
3Strong derandomization is sometimes impossible
- Setting Communication complexity.
- x(x1,x2) where x1,x2 are shared between two
players. - Exist randomized algorithms A(x,r) (e.g. for
Equality) with logarithmic communication
complexity s.t. any deterministic algorithm B(x)
requires linear communication. - Impossible to derandomize while preserving
complexity.
4(The easy direction of) Yaos Lemma A straight
forward averaging argument
- Given randomized algorithm that computes a
function f with success 1-? on the worst case,
namely - Given A0,1n0,1m ?0,1 s.t.
- ?x PrR?UmA(x,R)f(x)1-?
- ?r20,1m s.t. the deterministic algorithm
- B(x)A(x,r)
- computes f well on average, namely
- PrX?UnB(X)f(X)1-?
- Useful tool in bounding randomized algs.
- Can also be viewed as weak derandomization.
5Yaos lemma as weak derandomization
- Advantages
- Applies to any family of algorithms and any
complexity measure. - Communication complexity.
- Decision tree complexity.
- Circuit complexity classes.
- Construction B(x)A(x,r) preserves complexity.
- e.g. if A has low communication complexity then B
has low communication complexity.
- Drawbacks
- Weak derandomization is weak deterministic alg B
succeeds on most but not all inputs. - Lets not be too picky.
- In some scenarios (e.g. communication complexity)
strong derandomization is impossible. - The argument doesnt give an explicit way to find
r and produce B(x)A(x,r). - Uniformity is not preserved Even if A is
uniform we only get that B(x)A(x,r) is
nonuniform (B is a circuit).
6The goalExplicit versions of Yaos Lemma
- Given randomized algorithm that computes a
function f with success 1-? on the worst case,
namely - Given A0,1n0,1m ?0,1 s.t.
- ?x PrR?UmA(x,R)f(x)1-?
- Give explicit construction of a deterministic
algorithm B(x) s.t. - B computes f well on average, namely
- PrX?UnB(X)f(X)1-?
- Complexity is preserved complexity(B)
complexity(A). - We refer to this as Explicit weak
derandomization.
7Adelmans theorem (BPP?P/poly) follows from
Yaos lemma
- Given randomized algorithm A that computes a
function f with success 1-? on the worst case. - (amplification) amplify success prob. to 1-? for
?2-(n1) - (Yaos lemma) ? deterministic circuit B(x) such
that - b PrX?UnB(X)?f(X)lt?lt2-(n1) ? b0
- ? B succeeds on all inputs.
- Corollary Explicit version of Yaos lemma for
general poly-time algorithms ? BPPP. - Reminder of talk Explicit versions of Yaos
lemma for weak algorithms Communication games,
Decision trees, Streaming algorithms, AC0
algorithms.
8Related work Extracting randomness from the input
- Idea Goldreich and Wigderson Given a
randomized alg A(x,r) s.t. rx consider the
deterministic alg - B(x)A(x,x).
- Intuition If input x is chosen at random then
random coins rx is chosen at random. - Problem Input and coins are correlated.
(e.g. consider A s.t. 8input x, coin x is bad
for x). - GW Does work if A has the additional property
that whether or not a coin toss is good does not
depend on the input. - GW It turns out that there are As with this
property.
9The role of extractors in GW
- In their paper Goldreich and Wigderson actually
use - B(x)maj seeds y A(x,E(x,y))
- Where E(x,y) is a seeded extractor.
- Extractors are only used for deterministic
amplification (that is to amplify success
probability). - Alternative view of the argument
- Set A(x,r)maj seeds y A(x,E(r,y))
- Apply construction B(x)A(x,x).
10Randomness extractors
Do we have to tell that same old story again?
Daddy, how do computers get random bits?
11Randomness Extractors Definition and two flavors
- C is a class of distributions over n bit strings
containing k bits of (min)-entropy. - A deterministic (seedless) C-extractor is a
function E such that for every X?C, E(X) is
e-close to uniform on m bits. - A seeded extractor has an additional (short i.e.
log n) independent random seed as input. - For Seeded extractors Call X with
min-entropy k
source distribution from C
Seeded
Deterministic
- A distribution X has min-entropy k if
?x PrXx 2-k - Two distributions are e-close if the probability
they assign to any event differs by at most e.
12Zimand explicit version of Yaos lemma for
decision trees
- Zimand defines and constructs a stronger variant
of seeded extractors E(x,y) called exposure
resilient extractors . He considers - B(x)maj seeds y A(x,E(x,y))
- Thm Zimand07 If A is a randomized decision
tree with q queries that tosses q random coins
then - B succeeds on most inputs. (a (1-?)-fraction).
- B can be implemented by a deterministic decision
tree with qO(1) queries. - Zimand states his result a bit differently.
We improve to O(q)
13Our results
- Develop a general technique to prove explicit
versions of Yaos Lemma (that is weak
derandomization results). - Use deterministic (seedless) extractors that is
B(x)A(x,E(x)) where E is a seedless extractor. - The technique applies to any class of algorithms
with rx. Can sometimes handle rgtx using
PRGs. - More precisely Every class of randomized
algorithm defines a class C of distributions. An
explicit construction of an extractor for C
immediately gives an explicit version of Yaos
Lemma (as long as rx).
14Explicit version of Yaos lemma for communication
games
- Thm If A is a randomized (public coin)
communication game with communication complexity
q that tosses mltn random coins then set
B(x)A(x,E(x)) where E is a 2-source extractor. - B succeeds on most inputs.
A (1-?)-fraction (or even
a (1-2-(mq))-fraction). - B can be implemented by a deterministic
communication game with communication complexity
O(mq). - Dfn A communication game is explicit if each
party can compute its next message in poly-time
(given his input, history and random coins). - Cor Given an explicit randomized communication
game with complexity q and m coins there is an
explicit deterministic communicaion game with
complexity O(mq) that succeeds on a (1-2-(mq))
fraction of the inputs.
Both complexity and uniformity are preserved
15Explicit weak derandomization results
Extractors Algorithms
Extractors for bit-fixing sources KZ03,GRS04,R07 Decision trees (Improved alternative proof of Zimands result).
2-source extractors CG88,Bou06 Communication games
Construct from 2-source extractors. Inspired by KM06,KRVZ06. Streaming algorithms (can handle r x).
We construct inspired by PRGs for AC0. N,NW AC0 (constant depth) (can handle r x).
We construct using low-end hardness assumptions. Poly-time algorithms (can handle r x).
16Constant depth algorithms
- Consider randomized algorithms A(x,r) that are
computable by uniform families of poly-size
constant depth circuits. - NW,K Strong derandomization in quasi-poly
time. Namely, there is a uniform family of
quasi-poly-size circuits that succeed on all
inputs. - Our result Weak derandomization in poly-time.
Namely, there is a uniform family
of poly-size circuits that succeed on most
inputs. (can also preserve constant depth). - High level idea
- Reduce of random coins of A from nc to (log
n)O(1) using a PRG. (Based on the hardness of the
parity function H,RS) - Extract random coins from input x using an
extractor for sources recognizable by AC0
circuits. - Construct extractors using the hardness of the
parity function and ideas from NW,TV.
17High level overview of the proof
- To be concrete we consider communication games
18Preparations
- Thm If A is a randomized communication game with
communication complexity q that tosses m random
coins then set B(x)A(x,E(x)) where E is a
2-source extractor. - B succeeds on most inputs.
A (1-?)-fraction. - B can be implemented by a deterministic
communication game with communication complexity
O(mq). - Define independent random variables X,R by X?Un,
R?Um. - We have that ?x PrA(x,R)f(x)1-?
- It follows that a PrA(X,R)f(X)1-?
- We need to show b PrA(X,E(X))f(X)1-?
(2?2-2m). - The plan is to show that b a (2? 2-2m).
19High level intuition
x2
- For every choice of random coins r the
game A(,r) is deterministic w/complexity q. - It divides the set of strings x of length n into
2q rectangles. - Let Qr(x) denote the rectangle of x.
x1
Qr(x1,x2)
- At the end of protocol all inputs in a rectangle
answer the same way. - Consider the entropy in the variable Xrectangle
(XQr(X)v). - Independent of answer.
- Idea extract the randomness from this entropy.
- Doesnt make sense rectangle is defined only
after random coins r are fixed.
Rectangle 2-source
20Averaging over random coins and rectangles
x2
- For every choice of random coins r the
game A(,r) is deterministic w/complexity q. - It divides the set of strings x of length n into
2q rectangles. - Let Qr(x) denote the rectangle of x.
x1
Qr(x1,x2)
- a PrA(X,R)f(X)
- Sr PrA(X,R)f(X) ? Rr
- Sr Sv PrA(X,R)f(X) ? Rr ? Qr(X)v
- Sr Sv PrA(X,r) f(X) ? Rr ? Qr(X)v
- Sr Sv PrQr(X)vPrRrQr(X)vPrA(X,r)f(
X)Rr?Qr(X)v
21Averaging over random coins and rectangles
x2
- For every choice of random coins r the
game A(,r) is deterministic w/complexity q. - It divides the set of strings x of length n into
2q rectangles. - Let Qr(x) denote the rectangle of x.
x1
Qr(x1,x2)
- b PrA(X,E(X))f(X)
- Sr PrA(X, E(X))f(X) ? E(X)r
- Sr Sv PrA(X,E(X))f(X) ? E(X)r ? Qr(X)v
- Sr Sv PrA(X,r) f(X) ? E(X)r ? Qr(X)v
- SrSvPrQr(X)vPrE(X)rQr(X)v
PrA(X,r)f(X)E(X)r?Qr(X)v
22Proof (continued)
a PrA(X,R)f(X)
SrSvPrQr(X)vPrRr Qr(X)vPrA(X,r)f(X)
Rr ?Qr(X)v
SrSvPrQr(X)vPrE(X)rQr(X)vPrA(X,r)f(X)
E(X)r?Qr(X)v
b PrA(X,E(X))f(X)
23Proof (continued)
Would be fine if f was also constant over
rectangle
a PrA(X,R)f(X)
SrSvPrQr(X)vPrRr Qr(X)vPrA(X,r)f(X)
Rr ?Qr(X)v
v
2-m
x2
Problem It could be that A(,r) does well on
rectangle but poorly on E(X)r Note A(,r) is
constant over rectangle.
R is uniform and independent of X
x1
E is a 2-source extractor and Qr(X)v is a
rectangle
E(X)r
v
2-m
SrSvPrQr(X)vPrE(X)rQr(X)vPrA(X,r)f(X)
E(X)r?Qr(X)v
b PrA(X,E(X))f(X)
24Modifying the argument
- We have that PrA(X,R)f(x)1-?
- By Yaos lemma ?deterministic game F w/complexity
q - PrF(X)f(X)1-?
- Consider randomized algorithm A(x,r) which
- Simulates A(x,r)
- Simulates F(x)
- Let Qr(x) denote the rectangle of A and note
that - A(,r) is constant on rectangle Qr(X)v.
- F(x) is constant on rectangle Qr(X)v.
25Proof (continued)
Would be fine if f was also constant over
rectangle
a PrA(X,R)f(X)
SrSvPrQr(X)vPrRr Qr(X)vPrA(X,r)f(X)
Rr ?Qr(X)v
v
2-m
x2
Problem It could be that A(,r) does well on
rectangle but poorly on E(X)r Note A(,r) is
constant over rectangle.
R is uniform and independent of X
x1
E is a 2-source extractor and Qr(X)v is a
rectangle
E(X)r
v
2-m
SrSvPrQr(X)vPrE(X)rQr(X)vPrA(X,r)f(X)
E(X)r?Qr(X)v
b PrA(X,E(X))f(X)
26Proof (replace f?F)
We have that F is constant over rectangle!
a PrA(X,R)F(X)
a-a ?
SrSvPrQr(X)vPrRr Qr(X)vPrA(X,r)F(X)
Rr ?Qr(X)v
v
2-m
x2
Problem It could be that A(,r) does well on
rectangle but poorly on E(X)r Note A(,r) is
constant over rectangle.
R is uniform and independent of X
x1
E is a 2-source extractor and Qr(X)v is a
rectangle
E(X)r
v
2-m
SrSvPrQr(X)vPrE(X)rQr(X)vPrA(X,r)F(X)
E(X)r?Qr(X)v
b PrA(X,E(X))F(X)
b-b ?
27Finishing up
- Thm If A is a randomized communication game with
communication complexity q that tosses m random
coins then set B(x)A(x,E(x)) where E is a
2-source extractor. - B succeeds on most inputs.
A (1-?)-fraction. - B can be implemented by a deterministic
communication game with communication complexity
O(mq). - 2-source extractors cannot be computed by
communication games. - However, we need extractors for relatively large
rectangles. Namely 2-source extractors for
min-entropy n-(mq). - Each of the two parties can send the first 3(mq)
bits of his input. The sent strings have entropy
rate at least ½. - Run explicit 2-source extractor on substrings.
q.e.d.
???
28Generalizing the argument
- Consider e.g. randomized decision trees A(x,r).
- Define Qr(x) to be the leaf the decision tree
A(,r) reaches when reading x. - Simply repeat argument noting that Qr(X)v is a
bit-fixing source. - More generally, for any class of randomized
algorithms we can set Qr(x)A(x,r) - Can do the argument if we can explicitly
construct extractors for distributions that are
uniform over Qr(X)v
A(X,r)v. - Loosely speaking, need extractors for sources
recognizable by functions of the form A(,r). - There is a generic way to construct them from a
function that cannot be approximated by functions
of the form A(,r).
29Conclusion and open problem
- Loosely speaking Whenever we have a function
that is hard on average against a nonuniform
version of a computational model we get an
explicit version of Yaos lemma (that is explicit
weak derandomization) for the model. - Can handle AC0 using the hardness of parity.
- Gives a conditional weak derandomization for
general poly-time algorithms. Assumption is
incomparable to NW,GW. - Open problems
- Other ways to handle r gt x.
- Distributions that arent uniform.
30Thats it