Download presentation
Presentation is loading. Please wait.
Published byΞΞ±ΟΞΈΞ±Ξ―ΞΏΟ ΞΞ½Ξ±Ξ³Ξ½ΟΟΟΞΏΟ Modified over 6 years ago
1
Local Decoding and Testing Polynomials over Grids
Madhu Sudan Harvard University Joint work with Srikanth Srinivasan (IIT Bombay) January 11, 2018 ITCS: Polynomials over Grids
2
DeMillo-Lipton-Schwarz-Zippel
Notation: π½ = field, π½(π,π) = set of deg. β€π, π var. poly over π½ πΏ π,π = normalized Hamming distance πΏ π π = min πβπ½ π,π πΏ(π,π) DLSZ Lemma: Let πβπ½, with 2β€ π <β. If π,π: π π βπ½ satisfy πβ πβπ½ π,π then πΏ π,π β₯ 2 βπ . Strengthens to πΏ π,π β₯1βπ/|π| if π< π Holds even if π = {0,1} For this talk: π = constant DLSZ Lemma converts polynomials into error-correcting codes. With bonus: decodability, locality β¦ so πΏ π,π bounded away from 0. January 11, 2018 ITCS: Polynomials over Grids
3
Local Decoding and Testing
Informally: Given oracle access to π: π π βπ½ β¦ Testing: determine if πΏ π π β€π Decoding: If πΏ π π β€ πΏ 0 and π is nearest poly then determine value of π(π) given πβ π π Parameters: Query complexity = # of oracle queries Testing accuracy β‘ Pr rejection πΏ π π Decoding distance ( πΏ 0 above) Ideal: Query complexity π π (1), accuracy, dec. distance Ξ© π (1) Terminology π½(π,π) testable (decodable) over π if ideal achievable. Thms: For finite π½, π½(π,π) decodable over π½ (folklore) and testable over π½ ([Rubinfeld+S.β96] β¦. [Kaufman+Ronβ04]) January 11, 2018 ITCS: Polynomials over Grids
4
Polynomials over Grids
Testing + decoding thms hold only for π: π½ π βπ½ Not surprising β¦ repeated occurrence with polynomials. Even classical decoding algorithms (non-local) work only in this case! [Kopparty-Kimβ16]: Non-local Decoder (up to half the distance) for π: π π βπ½ for all finite π This talk: Testing + decoding when π= 0,1 βgridβ (Note: equivalent to π 1 Γβ―Γ π π with π π =2) January 11, 2018 ITCS: Polynomials over Grids
5
Obstacles to decoding/testing
Standard insight behind testing/decoding π< π½ π has deg. β€π iff π restricted to lines has β€π Can test/decode function on lines. Reduces complexity from π½ π to π½. When π> π½ use subspaces (= collection of linear restrictions) βAffine invarianceβ β β2 transitivityβ β¦ Problem over grids: General lines donβt stay within grid! Only linear restrictions that stay in grid: π₯ π =0 or π₯ π =1 π₯ π = π₯ π or π₯ π =1 β π₯ π (denoted β π₯ π = π₯ π βπβ) January 11, 2018 ITCS: Polynomials over Grids
6
ITCS: Polynomials over Grids
Main Results Thm 1. π½(1,π) not locally decodable over grid if char(π½) ββ No 2-transitivity! Serious obstacle! Thm 2. π½(π,π) is locally decodable over grid if char π½ <β. Query complexity = 2 O(max π,char π½ ) Decoding without 2-transitivity! Thm 3. π½(π,π) is locally testable over grid (βπ½) Testing without decoding! January 11, 2018 ITCS: Polynomials over Grids
7
ITCS: Polynomials over Grids
Proofs: Not LDC over β Idea π=1 : Consider π = linear function that is erased on βimbalancedβ points (of weight β π 2 βπ π , π 2 +π π ) Claim 1: No local constraints over β on balanced points and 0 π Claim 1.1: if π 0 π = π=1 β π π π π₯ π βπ then 0 π = π=1 β π π π₯ π Claim 1.2: 1β€ π π β€β! Claim 1.3: If π₯ 1 ,β¦, π₯ β are π-balanced then π π π π π₯ π is πβ
π π π π -balanced Claim 2: No local constraints β Not LDC Known over finite fields [BHRβ04] Not immediate for β January 11, 2018 ITCS: Polynomials over Grids
8
Proofs: LDC over small characteristic
Given: π πΏ-close to πβπ½(π,π), πβ 0,1 π (char π½ =π) Pick π: π β π uniformly and let π₯ π = π π β π¦ π(π) So considering π β² π¦ 1 ,β¦, π¦ π =π πβ π¦ π : π β² 0 =π π ; π β² 0 =π(π) Claim 1: For balanced π¦, have πβ π¦ π β₯π Claim 2: If π 2 = π π‘ and π 2 >π then πβ²(0) determined by πβ²(π¦) | π¦ = π 2 (Usual ingredient βLucasβs Theoremβ π₯ π + π¦ π = π₯+π¦ π ) January 11, 2018 ITCS: Polynomials over Grids
9
Proofs: LTC over all fields: Test
Pick πβ 0,1 π uniformly and π: π β π βrandomlyβ Verify π β² π¦ 1 ,β¦, π¦ π =π πβ π¦ π is of degree π Randomly = ? Not uniformly (donβt know how to analyze). Instead βrandom union-findβ Initially each π₯ π in πth bucket. Iterate: pick two uniformly random buckets and merge till π buckets left. Assign π¦ 1 ,β¦, π¦ π to π buckets randomly. Allows for proof by induction January 11, 2018 ITCS: Polynomials over Grids
10
Proofs: LTC over all field: Analysis
Key ingredients: (mimics BKSSZ analysis for RM) Main claim: Pr test rejects π β₯ 2 βπ(π) β
min 1, πΏ π (π) Proof by induction on π with 3 cases: Case 1: π moderately close to deg. π Show that eventually π disagrees from nearby poly on exactly one of 2 π samples. Usual proof β pairwise independence Our proof β hypercontractivity of sphere [Polyanskiy] + prob. fact about random union-find Case 2: Pr π,π,π π β π₯ π = π₯ π βπ very close to deg. π small Case 3: π far from deg. π, but Pr π,π,π β¦ high Use algebra to get contradiction. (Stitch different polynomials from diff. (π,π,π)βs to get nearby poly to π) Induction January 11, 2018 ITCS: Polynomials over Grids
11
Conclusions + Open questions
Many aspects of polynomials well-understood only over domain = π½ π Grid setting ( π π ) far less understood. When π={0,1} (equivalently π =2) testing possible without local decoding! (novel in context of low-degree tests!) General π open!! (even π =3 or even π= 0,1,2 ?) Is there a gap between characterizability and testability here? January 11, 2018 ITCS: Polynomials over Grids
12
ITCS: Polynomials over Grids
Thank You! January 11, 2018 ITCS: Polynomials over Grids
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.