quals-talk1

advertisement
Tolerant Locally Testable Codes
Atri Rudra
Qualifying Evaluation Project Presentation
Advisor: Venkatesan Guruswami
1
Fake Motivation

Elvis Presley is alive!


Check DNA


Verify this
Too much work
“Spot Check”



Accept Elvis
Reject Atri
Bruce Campbell ?
2
Outline of the talk







Real Motivation
Testing Codes
Previous work
Our Contributions
High Level ideas
Some Details
Open problems
3
Error Correcting Codes
C(x)
x
Encoder
y
Tester
Decoder
x
Hopeless
C(x)
x
Give up
4
Property testing
x


Verify a property
Oracle access to input



Does x have the property ?
Make few queries
Probabilistic tester


T
Accepts correct inputs
Rejects very bad inputs (whp)
0/1
5
Codes

Mapping C : k!n

Distance d = min u,v2  (C(u),C(v))
k



(¢,¢) is Hamming Distance
Rate k/n
[n,k,d]
d/2
d/2
d
6
Testing Codes
x




Property x 2? C
Make few queries
Probabilistic Tester
How good is the tester ?


Accept x 2 C w.p. 1
Reject x far from C w.p. 2/3


T
Hamming Distance
Local tester


Constant number of queries
Sub-linear also interesting
0 w.p.
1 2/3
7
Locally Testable Codes


Who Cares ?
Heart of PCPs


Alternate Characterization of NP
X 2? L





Proof (x)
Verifier checks (x)
Makes q queries
NP = PCP[ O(log n), O(1)]
[ALMSS92]…..
8
Another motivation
C(x)
x
y
x
Close
Far
Give up
9
Current Local Testers


Reject if y is far
Accept if y is close


By defn accepts only y2 C
Against rationale of codes
Close
y
Far
10
Tolerant Local Testers

Dist(y,C) <= c1d/n





y
Dist(y,C) > c2d/n


Accept w.p >= 2/3
Tolerance
Reject w.p. >= 2/3
Soundness
Close
Far
q(n) queries
(c1,c2,q)- testable


Prev work (0,O(1),O(1))-testable
Perfect completeness
11
The Holy Grail




Constant rate, linear distance
Constant Query Complexity
Not known even for LTCs
Unique decoding radius

c1=1/2, c2 ¼ 1/2?
d/2
d/2
d
12
Contributions

LTCs ! tolerant LTCs


Constant rate



Sub-linear query complexity
[BS04]
Constant # queries



No generic “complier”
Slightly Sub-constant rate
[BGHSV04]
Constant c1, c2
13
More on Contributions
(Constant # queries, Constant Rate)
Near uniform queries
Sub-constant Rate
Partitioned queries
Sub-linear # queries
Goal: Design codes and tolerant testers
14
Where are we now ?







Real Motivation
Testing Codes
Previous work
Our Contributions
High Level ideas
Some Details
Open problems
15
LTC ! tolerant LTC


Perfect Completeness
Uniform query pattern



x
c1= O(1/q) by union bound
Almost uniform is
q is not constant ?
T
1
16
Local Tester Revisited
x



Decision procedure is strict
Accept perturbations
There is a problem



Local View
Locally approx correct )
Global approx correct
Robustness

[BS04]
T
1
0
17
What is next ?



Constant rate, linear distance
Sub-linear query complexity
Product of Codes

[BS04]
18
Product of Codes


C [n,k,d]
C2




Any row 2 C
Any Column 2 C
[n2,k2,d2]
Tester ?
2C
n
C3
n
19
Tester for
2
C
row




pick row or clm
pick j2[n]
Rj2 C ?
Not known to be robust




Big open question
True for special cases
C is Reed-Solomon
C is C’2
n
C3?
n
20
Larger product of Codes


3
(C )
Similar definition (3D instead of 2D)
Same test



2? C2 test
Check all n2 pts
N2/3 queries


N=n3
2
2
2?2CC
2 C2
Robust!

[BS04]
21
Formal definition of Robustness

v2n




r random coin
T(v,r)=miny:T(y_r)=1 dist(v,y)
T(v)=Er[T(v,r)]
T is e-robust

8 v2n, dist(v,C)· e¢T(v)
22
3
C

Tolerant test



is tolerant LTC
Restriction is close to C2?
Constant rate
N2/3 queries



Reduce the # queries
Ct (t-Dimension)
N2/t queries
¼? C2
23
Tolerance of

dist(v,C)· 






tester
n3/3
h
f2 C3 closest to v
¸ 2n/3 choices of h


3
C
Dist(vh,fh)·  n2
Averaging argument
If not, for ¸ n/3 h, dist(vh,fh) >  n2
) dist(v,f)> n3/3
dist(vh,C2)·? n2
Similar arguments for other planes
v accepted w.p. ¸ 2/3
24
So what do we have now ?


Constant rate, linear distance
Sublinear query complexity



n # queries
 =2/t
C has no local tester but Ct has one
25
What is next ?

Slightly sub-constant rate, linear distance



n=k¢ exp(logk) for any >0
Constant query complexity
Based on PCPs

[BGHSV04]
26
PCP of Proximity



Variant of PCP introduced in [BGHSV04]
CKT-VAL(T)={x:T(x)=1}
Verifier VT such that




||=s¢ exp(logs)


x2 CKT-VAL(T), 9 , VT(x,)=1 wp 1
x far from CKT-VAL(T), 8 , VT(x,)=1 wp <1/2
#queries in hx,i
9 
8
x
s=|T|
Constant # queries
T
VT
1
0
27
Local Tester 1.0

Start with good code C0



Use PCPP as an aid


Constant rate and linear distance
Linear size encoding circuit
C1(x)= hC0(x),(x)i
There is a problem


|x|/|(x)|=o(1)
Distance of C1 is bad
(x)
x
x
(x)
C0
1
0
28
Local Tester 1.1

Increase the “code” part





C2(x)=h (C0(x))t,(x) i
Choose t such that |(x)|/(t¢|x|)=o(1)
Constant query complexity
Slightly sub-constant rate, linear distance
Not tolerant


Just corrupt the proof part
Corrupted word still close to C2
(C0(x))t
(x)
29
Tolerant Local Tester 1.2

Keep the code and proof parts comparable



C3(x)=h(C0(x))k,((x))li
k¢|C0(x)|=(l¢|(x)|)
Need near uniform queries

Constant query complexity
Slightly sub-constant rate, Linear distance

Used in relaxed LDC in [BGHSV04]

30
To summarize


Defined tolerant LTCs
Explicit constructions



Constant # queries, slightly sub-constant rate
Sub-linear # queries, constant rate
Both constructions start from some C0

C0 does not have a (tolerant) local tester
31
Open Questions

Is “natural” tester for C2 robust ?


e-robust for e=O(1)
row
No lower bounds on n for LTCs

Does tolerance make lower bounds easier ?
n
C3?
n
32
Questions ?
33
Download