1 Tolerant Locally Testable Codes Atri Rudra Qualifying Evaluation Project Presentation Advisor: Venkatesan Guruswami
2 Fake Motivation Elvis Presley is alive! Verify this Check DNA Too much work “Spot Check” Accept Elvis Reject Atri Bruce Campbell ?
3 Outline of the talk Real Motivation Testing Codes Previous work Our Contributions High Level ideas Some Details Open problems
4 Error Correcting Codes x Encoder C(x) Decoder xGive up y x C(x) Tester Hopeless
5 Property testing Verify a property Oracle access to input Does x have the property ? Make few queries Probabilistic tester Accepts correct inputs Rejects very bad inputs (whp) x T 0/1
6 Codes Mapping C : k ! n Distance d = min u,v2 k (C(u),C(v)) ( ¢,¢) is Hamming Distance Rate k/n [n,k,d] d/2 d
7 Testing Codes 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 Hamming Distance Local tester Constant number of queries Sub-linear also interesting T 1 x 0 w.p. 2/3
8 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]…..
9 Another motivation C(x) x y x Give up Far Close
10 Current Local Testers Reject if y is far Accept if y is close By defn accepts only y2 C Against rationale of codes y Far Close
11 Tolerant Local Testers Dist(y,C) <= c 1 d/n Accept w.p >= 2/3 Tolerance Dist(y,C) > c 2 d/n Reject w.p. >= 2/3 Soundness q(n) queries (c 1,c 2,q)- testable Prev work (0,O(1),O(1))-testable Perfect completeness y Far Close
12 The Holy Grail Constant rate, linear distance Constant Query Complexity Not known even for LTCs Unique decoding radius c 1 =1/2, c 2 ¼ 1/2? d/2 d
13 Contributions LTCs ! tolerant LTCs No generic “complier” Constant rate Sub-linear query complexity [BS04] Constant # queries Slightly Sub-constant rate [BGHSV04] Constant c 1, c 2
14 More on Contributions (Constant # queries, Constant Rate) Sub-constant Rate Sub-linear # queries Near uniform queries Partitioned queries Goal: Design codes and tolerant testers
15 Where are we now ? Real Motivation Testing Codes Previous work Our Contributions High Level ideas Some Details Open problems
16 LTC ! tolerant LTC Perfect Completeness Uniform query pattern c 1 = O(1/q) by union bound Almost uniform is q is not constant ? x T 1
17 Local Tester Revisited Decision procedure is strict Accept perturbations There is a problem Local View Locally approx correct ) Global approx correct Robustness [BS04] 0 x T 1
18 What is next ? Constant rate, linear distance Sub-linear query complexity Product of Codes [BS04]
19 Product of Codes C [n,k,d] C 2 Any row 2 C Any Column 2 C [n 2,k 2,d 2 ] Tester ? n n C3C3 2 C
20 Tester for C 2 pick row or clm pick j2[n] R j 2 C ? Not known to be robust Big open question True for special cases C is Reed-Solomon C is C’ 2 n n C3?C3? row
21 Larger product of Codes (C 3 ) Similar definition (3D instead of 2D) Same test 2 ? C 2 test Check all n 2 pts N 2/3 queries N=n 3 Robust! [BS04] 2 C 2 2 ? C 2
22 Formal definition of Robustness v2 n r random coin T (v,r)=min y:T(y_r)=1 dist(v,y) T (v)=E r [ T (v,r)] T is e-robust 8 v2 n, dist(v,C)· e¢ T (v)
23 C 3 is tolerant LTC Tolerant test Restriction is close to C 2 ? Constant rate N 2/3 queries Reduce the # queries C t (t-Dimension) N 2/t queries ¼ ? C 2
24 Tolerance of C 3 tester dist(v,C)· n 3 /3 f2 C 3 closest to v ¸ 2n/3 choices of h Dist(v h,f h )· n 2 Averaging argument If not, for ¸ n/3 h, dist(v h,f h ) > n 2 ) dist(v,f)> n 3 /3 Similar arguments for other planes v accepted w.p. ¸ 2/3 dist(v h,C 2 )· ? n 2 h
25 So what do we have now ? Constant rate, linear distance Sublinear query complexity n # queries =2/t C has no local tester but C t has one
26 What is next ? Slightly sub-constant rate, linear distance n=k¢ exp(log k) for any >0 Constant query complexity Based on PCPs [BGHSV04]
27 PCP of Proximity Variant of PCP introduced in [BGHSV04] CKT-VAL(T)={x:T(x)=1} Verifier V T such that x2 CKT-VAL(T), 9 V T (x, )=1 wp 1 x far from CKT-VAL(T), 8 V T (x, )=1 wp <1/2 #queries in hx, i | |=s¢ exp(log s) s=|T| Constant # queries VTVT x T
28 Local Tester 1.0 Start with good code C 0 Constant rate and linear distance Linear size encoding circuit Use PCPP as an aid C 1 (x)= hC 0 (x), (x)i There is a problem |x|/| (x)|=o(1) Distance of C 1 is bad C0C0 x x) 1 0 x
29 Local Tester 1.1 Increase the “code” part C 2 (x)=h (C 0 (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 C 2 (C 0 (x)) t x)
30 Tolerant Local Tester 1.2 Keep the code and proof parts comparable C 3 (x)=h(C 0 (x)) k,( (x)) l i k¢|C 0 (x)|= (l¢| (x)|) Need near uniform queries Constant query complexity Slightly sub-constant rate, Linear distance Used in relaxed LDC in [BGHSV04]
31 To summarize Defined tolerant LTCs Explicit constructions Constant # queries, slightly sub-constant rate Sub-linear # queries, constant rate Both constructions start from some C 0 C 0 does not have a (tolerant) local tester
32 Open Questions Is “natural” tester for C 2 robust ? e-robust for e=O(1) No lower bounds on n for LTCs Does tolerance make lower bounds easier ? n n C3?C3? row
33 Questions ?