340 likes | 484 Views
This lecture delves into the critical aspects of error detection and correction within coding theory, highlighting key algorithms and the implications of sublinear time computation. It explores concepts like Locally Testable Codes (LTCs) and Locally Decodable Codes (LDCs), their construction, and applications in various contexts such as property testing and interactive proofs. The session aims to provide foundational definitions, examine known results, and discuss novel approaches in error correction, emphasizing the intersections of coding theory and algorithmic efficiency.
E N D
Local Error-Detection and Error-correction Madhu Sudan MIT Coding & Sublinear time
k k k k k ( ) ( ( ( ) ) ) d d d d d d d f f 9 9 h ± G F G E G C § C § § § § I E E § C § § § § § E E i i i i n n n n i i i i i n i n i i µ · t t t t t t 2 2 2 2 2 2 x v ² e - n o v v : x e e n n e x x a n c o ; a m s m s p e e o u a c c c g e e a e e m e m m : a s s m ; n x m z e m s m x ² ! = ! = , , , . . . . . ; . . k ( ( ( ) ) ) ( = ( ) ( ( ( ) ) ) ) ± k ± d d ± l d d G E R C § C E E i i i i · t t 2 v e n m m x p r n o c v o m e p u e n o r m m m a x z e ² s a n c e = = ; , , ; . . . Algorithmic Problems in Coding Theory • Code: • Encoding: • Error-detection ( Testing): • Error-correction (Decoding): Coding & Sublinear time
l x - o r a c e 0 ( ) f k x ( ( f ) ) g f g i f f k n i j 0 1 0 1 x o x : x n ! j ; ; ; 0 ( ) h f ¼ w e r e x x x i f Sublinear time algorithmics • Given can it be “computed” in time? • Answer 1: Clearly NO, since that is the time it takes to even read the input/write the output • Answer 2: YES, if we are willing to • Present input implicitly (by an oracle). • Represent output implicitly • Compute function on approximation to input. Extends to computing relations as well. Coding & Sublinear time
Sub-linear time algorithms • Initiated in late eighties in context of • Program checking • Interactive Proofs/PCPs • Now successful in many more contexts • Property testing/Graph-theoretic algorithms • Sorting/Searching • Statistics/Entropy computations • (High-dim.) Computational geometry • Many initial results are coding-theoretic! Coding & Sublinear time
Sub-linear time algorithms & Coding • Encoding: Not reasonable to expect in sub-linear time. • Testing? Decoding? – Can be done in sublinear time. • In fact many initial results do so! • Codes that admit efficient … • … testing: Locally Testable Codes (LTCs) • … decoding: Locally Decodable Codes (LDCs). Coding & Sublinear time
Rest of this talk • Definitions of LDCs and LTCs • Quick description of known results • Some basic constructions • (Time permitting) Yekhanin’s construction of LDCs. Coding & Sublinear time
Definitions Coding & Sublinear time
n k ( ) ( ( ) ( ) = ) h f d ± ? d h l l d d f b l W C D § § C M L D i n i i i i i 2 t t t t > : a r e a ² s q n s r q a n ² - o m o c g p a o s y n e o e n e c s o o o a w e ! w ; [ ] = f 9 d d l f ` l d k d D D i i i i 2 3 t t t t t t t t 2 r a e n p e o c r o o u a e p r u s s s o m u w g p v p o e n a c o e e a w s o r s i . . . . . . ( ( ) ) ( ) = d l 9 ± ± C C 2 · · t a n o r a c e w s m w m ² . . ; , Locally Decodable Code Code: Coding & Sublinear time
n ( ( ( ) ) ) ( ) ` ` l l d l l d l d d b d l d f 8 b l f 9 f d C D C § D D i i i i i i i i n i j t t t t 2 s q s ² ² r - e a - o c s s a - q y n e c o s r a - a n e c e o o m a p w o e s o n s o e c o w e r s w ; ; ; ; , . . [ ] ( ) [ ] ( ) = # d d k d d ` ± d l l ` C i i i j 2 3 · t t t t t t t t 2 2 2 g v c e a o n n e w o o u r p s u a c n s m s w a p n w a o c r a e c a s e ² w s s a m o s j i . . . . ; . . . . ( ( ) ) l l f ± C i i · t m m a r e a m e s s a g e s s a s y n g w m ² ` j 1 ; : : : ; ; Locally List-Decodable Code Code: Coding & Sublinear time
History of definitions • Constructions predate formal definitions • [Goldreich-Levin ’89]. • [Beaver-Feigenbaum ’90, Lipton ’91]. • [Blum-Luby-Rubinfeld ’90]. • Hints at definition (in particular, interpretation in the context of error-correcting codes): [Babai-Fortnow-Levin-Szegedy ’91]. • Formal definitions • [S.-Trevisan-Vadhan ’99] (local list-decoding). • [Katz-Trevisan ’00] Coding & Sublinear time
n ( ( ) ) d d l l b l C T § L T n i i i µ t t r e a s q s n q r ² a n - o o c m a p y o s e s o n a s : e w ; f 9 f T I T C i 1 t t t 2 ² e s w e r s a c c e p s w p . . . . . = f f f h I C i j 1 2 ¸ t t ² w s ² - a r r o m e n r e e c s w p , . . . Locally Testable Codes Code: “Weak” definition: hinted at in [BFLS], explicit in [RS’96, Arora’94, Spielman’94, FS’95]. Coding & Sublinear time
n ( ( ) ) d d l l b l T C § L T n i i i µ t t r e a s q s n q r ² a n - o o c m a p y o s e s o n a s : e w ; f 9 f T I T C i 1 t t t 2 ² e s w e r s a c c e p s w p . . . . . F § n 2 ² o r e v e r y w , ( ( ) ) ± T C j ¸ t r e e c s w p w . . ; . Strong Locally Testable Codes Code: “Strong” Definition: [Goldreich-S. ’02] Coding & Sublinear time
Motivations Coding & Sublinear time
N ( ( ) ) l l d d f h f L A C i i i i i t t t t t t t t t l l l d d b l d f S C § N i n 2 µ f g ¼ ² ² b d f o c e a C r n a e c e o n n e g r p r e a c a o n n : c o m o p m u p u e c e x c x o r w e v e o r y u § x r e - i i n 0 1 ) t ² u p p o s e s o c a y - e c o a e c o e o r 2 = ² c c a n e v e w e a s u n c o n c : ! , . ; . ( 0 ) 0 [ ] l d f d l L I H R B F P i i i i ( t t t t t t h l l l d d b f h d d F i t t t o v e n a e c n a g n x c o m p e u a e s c x o n o s r a m n c o e s x n g e a e s a v e r r v a a g e e - u r e r a s s u m e c a n o c a y e c o e s o e c o e w o r . . , , [ ] [ ] f l l I R C G K S L S T V i i i i ) t t t t t t d b f h j i t t t t c n a s o e r c m o a m p o n e x e y r o e v w a o r s - c a s e p o n a n n o u s s o e m e s s a g e . . , . Motivations for Local decoding Coding & Sublinear time
Motivation for Local-testing • No generic applications known. • However, • Interesting phenomenon on its own. • Intangible connection to Probabilistically Checkable Proofs (PCPs). • Potentially good approach to understanding limitations of PCPs (though all resulting work has led to improvements). Coding & Sublinear time
Contrast between decoding and testing • Decoding: Property of words near codewords. • Testing: Property of words far from code. • Decoding: • Motivations happy with n = quasi-poly(k), and q = poly log n. • Lower bounds show q = O(1) and n = nearly-linear(k) impossible. • Testing: Better tradeoffs possible! Likely more useful in practice. • Even conceivable: n = O(k) with q = O(1)? Coding & Sublinear time
Some LDCs and LTCs Coding & Sublinear time
( = ) j j = j j k l ± f f d l l ± f l l F F F F F P P m i i m m i m i ¸ t t t t t t ¼ e c o v a e u a c e m o n n s s o o n e g o n a m - v o a r a e p o y n o m a = , , , . . ¯ ¯ l d F i t o v e r n e e P Codes via Multivariate Polynomials Message: Encoding: Parameters: (Reed Muller code) Coding & Sublinear time
k f f b l h d l h f d f f d d V P i i i i i i i i i t t t t t t t t t t t t t m e - c r v a y r s u a e s r p e p a s o c r e y c n o e r m o u a o g s o p p a o c e e n g r s e x e o o r e e n g s r e e r r e e c s e o . , 0 ( ) d d d d l ± b b l l S i i i i t < m a n a m e m e c c o - o m p e m o e x n a s y u n s e p a s c u e s p a c e s p o y n o m a . . . f d t o e g r e e 0 j j ( ) . l l F Q T i i m t u e r y c o m p e x y q ; m e p o y q = = . 0 b l ! i ¿ m m s u n e a r ) Basic insight to locality • Local Decoding: • Local Testing: Coding & Sublinear time
( ) k l k c ¢ o g = . = ( ) = ( ) l 1 1 1 ¡ ( ) ( ) ( ) l l l l d d b l b b h l l h h h k d k k L L L L T D D i i i i i i i i i i i q p o y q ² 3 t t t t t t t t t o o o o c c c c a a a a y e e e s q c c o o a w a a y n w y y w w e x p n q e x a p n n e x p = = = = ~ ( ) ( ) d k O O 1 q a n n = = Summary of Constructions • Polynomial Codes: (Locally decodable and testable) • Polynomial Codes + Composition/Concatenation: • Codes based on “Algebraic Designs” [Yekhanin] Coding & Sublinear time
[Yekhanin ’07]’s LDCs Coding & Sublinear time
( ( ( ( ) ) ) ) ( ) l l l l l k T L L S S £ i y m m a a p r r a a g g c e e a a n s w e r m = j ( j j j ) l f d d k f b ? ? S H S S A S T T T T i t \ \ o s w a a u r g n e e o c v c e a o n n n o m e k k i i i j 1 1 ; : : : ; , ; : : . . : ; . f g S T 1 µ m i i ; ; : : : ; . Recall: Combinatorial Designs • Families of Sets: • Restrictions on Intersections: • E.g., • Basic Question: i vs. i: i vs. j: Coding & Sublinear time
j j ¡ ¢ S 1 ¡ ( ) m l d l d ! B A S i i i p i t t p p a s s m m c a o - p s - e p m s m r e g s m n g e n » h h h h i i i i l 6 6 k b ? H S 0 0 0 0 1 ; ¡ 2 3 u p u u u u o w v v v v u a r v g e c a n v e = = = k k i i i i i i j j 1 1 ; : : : ; ; ; ; ; , ; : : : ; . h ? C i i t a n w e a c e v e F m 2 u v i i ; . p [Yekhanin]’s Algebraic Designs • Families of Vectors: • Restrictions on Inner Products: • Basic Question: Coding & Sublinear time
j j ¡ ¢ S 1 ¡ ( ) m l d l d ! B A S i i i p i t t p p a s s m m c a o - p s - e p m s m r e g s m n g e n » h h h h i i i i l 6 6 k b ? H S 0 0 0 0 1 ; ¡ 2 3 u p u u u u o w v v v v u a r v g e c a n v e = = = k k i i i i i i j j 1 1 ; : : : ; ; ; ; ; , ; : : : ; . h ? C i i t a n w e a c e v e F m 2 u v i i ; . p [Yekhanin]’s Algebraic Designs • Families of Vectors: • Restrictions on Inner Products: • Basic Question: Coding & Sublinear time
= 1 1 1 ¡ ¡ ( ) ( ) ( ) k l 9 b d k f h k ¤ F F L L S A B S S p i i i i i p i m 1 2 · µ t t t q e e p m m m m m a a - : : g q e a s r c a q p c n - p n e c s e e g n x n e p s w p s o s v e c o r s n = ) = = ; ; . . . p p ( ) ( ) d b h k k b b F S L D C i i i i i i i m m t t t t t p p - q - u e e r s y g n w n a r y v e c o s r s m a n p p n g - s o p s ) ; p k b b L D C i i i m t t t q - q u e r y s m a p p n g s o p s ) . [Y’07] Algebraic designs and LDCs (Matches some of the early constructions) Coding & Sublinear time
( ( ) ( ) ) ¯ f l 9 b l l b d ¯ l f l ¤ f F D L O S S S S i i i i i i i i i i i 2 · µ t t t t t q e e p m n e n m o a - o n a w : : g o e q e q r s u a q q c v - a a p n e g c n e e n r e a s e s c p a n o s y o n n s c e = ; ; . . . p ( ) ( ) [ ] = ( ) 9 d l h k l h F F S i i i i m p 1 t t t ¡ 2 p a q - - s p e a s r s g e n p w o y n o m v e a c o r s x n x x s 2 ; . . p ¯ f ( ) j g d l d b h ¯ k b b l L D C S i i i i i i i m t t t t t 2 e a q - q g u e e n r e y r a e s y m a p p x n g s o s n p o n - r s v a ) . . [Y’07] Algebraic designs and LDCs Coding & Sublinear time
= 7 1 7 ( ) ( ) ( ) f ( g ) l l l l b 9 b d l l k ! b f ¤ k b F S E L S L S D C S S S i i i i i i i i i i 3 3 2 1 2 7 1 2 4 8 1 6 3 2 6 4 · µ t t t t t m q e x p m a s m - m o q n - p u a a g - e e g r : a : y e p p g r e q a r - c a a e q c m s y ; p n a g n n p c p s e c n e e n e x g s s s p o s s o e x p s ) = = = ; ; ; ; ; . . ; ; ; . ; p ( ) d h k F S i i i m t t p - e s g n w v e c o r s n ; p k b b L D C i i i m t t t q - q u e r y s m a p p n g s o p s ) . [Y’07] Algebraic designs and LDCs Coding & Sublinear time
1 t t ¡ ( ) ( ) f g h l 9 9 b l l f b ¤ f F F T L L L S S S L D C S S i i i i i 2 3 4 3 2 1 1 2 4 2 · µ t t t ¡ q e e e p m m m e o m m m r e a a a m - a : : : : g p e q m - r q a u u q c e r p p n y c c e a n e s v s e p o s s u g r o u p o = = ) = p ; ; . . ; ; ; . : : : ; p j j S 0 0 0 0 0 0 0 1 ( ) ( ) ( ) 9 l d b k d b l l h k f l k h b F S S i i i i i i i i i i m 3 t t t t t t : m s p a p - a p - p g n e g e s r g a - n c e a w s s g y n o n o e c x e v p e e n c g o r s n m s ) » ; ; . . . p k b b L D C i i i m t t t q - q u e r y s m a p p n g s o p s ) . [Y’07] Algebraic designs and LDCs Coding & Sublinear time
Proofs? • Disclaimer: Proof of Lemma 2, Lemma 3 too long to fit here. (Many context switches, but elementary.) • Will only attempt to show Lemmas 1 and 4. Coding & Sublinear time
G i v e n u u ; v v k k 1 1 ; : : : ; ; : : : ; 2 3 u x i . . . 6 7 G 1 ¢ ¢ ¢ ¢ ¢ ¢ = h i 6 7 u x i ; 4 5 . . . Basic designs and LDCs Coding & Sublinear time
( ) 2 1 + + + ¡ y m y y v p v y m v k i i i 1 ; : : : ; h i 6 0 t s u y = i . . ; 2 3 . . . 6 7 1 ¢ ¢ ¢ ¢ ¢ ¢ h i 6 7 u x i ; 4 5 . . . Basic designs and LDCs message … codeword … Report parity of locations Coding & Sublinear time
h h i t t o e r r o w r o w s l l 1 ¡ - - p a o n e o s n e s . . 2 3 . . 2 3 . 6 7 . ? = . 6 7 . 4 5 6 7 1 . ¢ ¢ ¢ ¢ ¢ ¢ h i . 6 7 u x i . ; 4 5 . . . Basic designs and LDCs + + + … codeword Coding & Sublinear time
h i h = j j i j j f g k h h f b h l f f l f f U T S ~ ~ i i i i i i i j i i 0 1 1 1 t t t t t t t t t t t t t ¡ ¡ \ 2 v u u a s e e v u s c c e u a a v o r v r a a c s c ; a z e s p e e r r a u s e s x o a c v c v c e v v ; e e y c e c n p o o s r r o r o o p c o s o e w m e r p s e o m e u n p v o s e $ = = = = = i i i i i i i i i j i i ; ; ; ; . ; : : : ; ; . . Proof of Lemma 4 • Construction of Basic p-designs: • Construction of (p,S)-designs for S multiplicative: Coding & Sublinear time
Conclusions • Local algorithms in error-detection/correction lead to interesting new questions. • Non-trivial progress so far. • Limits largely unknown • O(1)-query LDCs must have R(C) = 0 [Katz-Trevisan] Coding & Sublinear time