Algebraic Property Testing

Slides:



Advertisements
Similar presentations
1 Property testing and learning on strings and trees Michel de Rougemont University Paris II & LRI Joint work with E. Fischer, Technion, F. Magniez, LRI.
Advertisements

Parikshit Gopalan Georgia Institute of Technology Atlanta, Georgia, USA.
December 2, 2009 IPAM: Invariance in Property Testing 1 Invariance in Property Testing Madhu Sudan Microsoft/MIT TexPoint fonts used in EMF. Read the TexPoint.
Grigory Yaroslavtsev Joint work with Piotr Berman and Sofya Raskhodnikova.
Of 22 August 29-30, 2011 Rabin ’80: APT 1 Invariance in Property Testing Madhu Sudan Microsoft Research TexPoint fonts used in EMF. Read the TexPoint manual.
Property testing of Tree Regular Languages Frédéric Magniez, LRI, CNRS Michel de Rougemont, LRI, University Paris II.
A UNIFIED FRAMEWORK FOR TESTING LINEAR-INVARIANT PROPERTIES ARNAB BHATTACHARYYA CSAIL, MIT (Joint work with ELENA GRIGORESCU and ASAF SHAPIRA)
Of 37 October 18, 2011 Invariance in Property Testing: Chicago 1 Invariance in Property Testing Madhu Sudan Microsoft Research TexPoint fonts used in EMF.
Of 39 February 22, 2012 Invariance in Property Testing: Yale 1 Invariance in Property Testing Madhu Sudan Microsoft Research TexPoint fonts used in EMF.
1 Markov Decision Processes: Approximate Equivalence Michel de Rougemont Université Paris II & LRI
Christian Sohler | Every Property of Hyperfinite Graphs is Testable Ilan Newman and Christian Sohler.
Artur Czumaj Dept of Computer Science & DIMAP University of Warwick Testing Expansion in Bounded Degree Graphs Joint work with Christian Sohler.
Proclaiming Dictators and Juntas or Testing Boolean Formulae Michal Parnas Dana Ron Alex Samorodnitsky.
Learning and Fourier Analysis Grigory Yaroslavtsev CIS 625: Computational Learning Theory.
Testing the Diameter of Graphs Michal Parnas Dana Ron.
Sparse Random Linear Codes are Locally Decodable and Testable Tali Kaufman (MIT) Joint work with Madhu Sudan (MIT)
Michael Bender - SUNY Stony Brook Dana Ron - Tel Aviv University Testing Acyclicity of Directed Graphs in Sublinear Time.
Testing Metric Properties Michal Parnas and Dana Ron.
On Proximity Oblivious Testing Oded Goldreich - Weizmann Institute of Science Dana Ron – Tel Aviv University.
1 Sampling Lower Bounds via Information Theory Ziv Bar-Yossef IBM Almaden.
On Testing Convexity and Submodularity Michal Parnas Dana Ron Ronitt Rubinfeld.
Quantum Algorithms II Andrew C. Yao Tsinghua University & Chinese U. of Hong Kong.
On Testing Computability by small Width OBDDs Oded Goldreich Weizmann Institute of Science.
Some 3CNF Properties are Hard to Test Eli Ben-Sasson Harvard & MIT Prahladh Harsha MIT Sofya Raskhodnikova MIT.
Of 29 12/09/2014 Invariance in Property 1 Two Decades of Property Testing Madhu Sudan Microsoft Research TexPoint fonts used in EMF. Read.
September 20, 2010 Invariance in Property Testing 1 Madhu Sudan Microsoft/MIT TexPoint fonts used in EMF. Read the TexPoint manual before you delete this.
Of 38 July 29, 2011 Invariance in Property Testing: EPFL 1 Invariance in Property Testing Madhu Sudan Microsoft Research TexPoint fonts used in EMF. Read.
March 20-24, 2010 Babai-Fest: Invariance in Property Testing 1 Invariance in Property Testing Madhu Sudan Microsoft/MIT TexPoint fonts used in EMF. Read.
January 8-10, 2010 ITCS: Invariance in Property Testing 1 Invariance in Property Testing Madhu Sudan Microsoft/MIT TexPoint fonts used in EMF. Read the.
Of 28 Probabilistically Checkable Proofs Madhu Sudan Microsoft Research June 11, 2015TIFR: Probabilistically Checkable Proofs1.
Approximating the MST Weight in Sublinear Time Bernard Chazelle (Princeton) Ronitt Rubinfeld (NEC) Luca Trevisan (U.C. Berkeley)
Of 29 August 4, 2015SIAM AAG: Algebraic Codes and Invariance1 Algebraic Codes and Invariance Madhu Sudan Microsoft Research.
The Quasi-Randomness of Hypergraph Cut Properties Asaf Shapira & Raphael Yuster.
Limits of Local Algorithms in Random Graphs
Succinct representation of codes with applications to testing Elena Grigorescu Tali Kaufman Madhu Sudan.
Transitive-Closure Spanner of Directed Graphs Kyomin Jung KAIST 2009 Combinatorics Workshop Joint work with Arnab Bhattacharyya MIT Elena Grigorescu MIT.
1 Approximate Schemas and Data Exchange Michel de Rougemont University Paris II & LRI Joint work with Adrien Vielleribière, University Paris-South.
COMPLEX ZEROS: FUNDAMENTAL THEOREM OF ALGEBRA Why do we have to know imaginary numbers?
Polynomials Emanuele Viola Columbia University work partially done at IAS and Harvard University December 2007.
Norms, XOR lemmas, and lower bounds for GF(2) polynomials and multiparty protocols Emanuele Viola, IAS (Work partially done during postdoc at Harvard)
狄彥吾 (Yen-Wu Ti) 華夏技術學院資訊工程系 Property Testing on Combinatorial Objects.
Complexity and Efficient Algorithms Group / Department of Computer Science Testing the Cluster Structure of Graphs Christian Sohler joint work with Artur.
Of 17 Limits of Local Algorithms in Random Graphs Madhu Sudan MSR Joint work with David Gamarnik (MIT) 7/11/2013Local Algorithms on Random Graphs1.
Of 29 April 8, 2016 Two Decades of Property Testing 1 Madhu Sudan Harvard TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.:
Complexity and Efficient Algorithms Group / Department of Computer Science Testing the Cluster Structure of Graphs Christian Sohler joint work with Artur.
Tali Kaufman (Bar-Ilan)
On Sample Based Testers
Algebraic Property Testing:
Property Testing (a.k.a. Sublinear Algorithms )
Cyclic Codes 1. Definition Linear:
Locality in Coding Theory II: LTCs
Approximating the MST Weight in Sublinear Time
Polynomial Norms Amir Ali Ahmadi (Princeton University) Georgina Hall
3.8 Complex Zeros; Fundamental Theorem of Algebra
Sum of Squares, Planted Clique, and Pseudo-Calibration
Algebraic Codes and Invariance
General Strong Polarization
Subhash Khot Dept of Computer Science NYU-Courant & Georgia Tech
Local Error-Detection and Error-correction
CIS 700: “algorithms for Big Data”
Locally Decodable Codes from Lifting
Approximate Validity of XML Streaming Data
Invariance in Property Testing
Beer Therapy Madhu Ralf Vote early, vote often
General Strong Polarization
Locality in Coding Theory II: LTCs
Universal Semantic Communication
General Strong Polarization
General Strong Polarization
Presentation transcript:

Algebraic Property Testing Madhu Sudan MIT CSAIL Joint work with Tali Kaufman (IAS).

Classical Data Processing Big computers Small Data

Modern Data Processing Small computers Enormous Data

Needs new algorithmic paradigm Imbalance not a question of technology : I.e., not because computing speeds are growing less fast than memory capacity. Imbalance is a function of expectations: E.g., Users expect to be able to “analyze” the WWW, using a laptop. But WWW includes millions other such laptops. Need: Sublinear time algorithms That “estimate” rather than “compute” some given function.

Property Testing D a t : f ! R f g i v e n b y a s m p l o x P r o p e f(x) P r o p e t y : F µ f D ! R g q - u e r y T s t : S a m p l f b o x i . A c e p t s i f 2 F Hope: I m p o s i b l e w t h q ¿ j D R e j c t s i f 6 2 F R e j c t s i f ± - a r o m F f ± - c l o s e t F i 9 g 2 . P r x D [ ( ) 6 = ] ·

Property Testing D a t : f ! R f g i v e n b y a s m p l o x P r o p e f(x) P r o p e t y : F µ f D ! R g q - u e r y T s t : S a m p l f b o x i . A c e p t s i f 2 F Hope: I m p o s i b l e w t h q ¿ j D R e j c t s i f 6 2 F R e j c t s i f ± - a r o m F f ± - c l o s e t F i 9 g 2 . P r x D [ ( ) 6 = ] ·

Example: Linearity Testing f : F n 2 ! g [Blum, Luby, Rubinfeld ’90] Property = Linearity Test: Non-trivial analysis: D = F n 2 ; R F = f j 8 x ; y ( ) + g P i c k r a n d o m x ; y A c e p t i f ( x ) + y = F f ± - a r o m F ) e j c t w . p 2 = 9

Example: Linearity Testing f : F n 2 ! g [Blum, Luby, Rubinfeld ’90] Property = Linearity Test: Non-trivial analysis: D = F n 2 ; R F = f j 8 x ; y ( ) + g P i c k r a n d o m x ; y A c e p t i f ( x ) + y = F f ± - a r o m F ) e j c t w . p 2 = 9

Property Testing: Abbreviated History Prehistoric: Statistical sampling E.g., “Is mean/median at least X”. Linearity Testing [BLR’90], Multilinearity Testing [Babai, Fortnow, Lund ’91]. Graph/Combinatorial Property Testing [Goldreich, Goldwasser, Ron ’94]. E.g., Is a graph “close” to being 3-colorable. Algebraic Testing [GLRSW,RS,FS,AKKLR,KR,JPSZ] Is multivariate function a polynomial (of bounded degree). Graph Testing [Alon-Shapira, AFNS, Borgs et al.] Characterizes graph properties that are testable.

This Talk I f F µ ! i s c l o e d u r a t , ± m h y z b . Abstracting Algebraic Property Testing Generic Theorem: Motivations: Generalizes, unifies previous algebraic works Initiates systematic study of testability for algebraic properties Sheds light on testing and invariances of properties. I f F µ n ! i s c l o e d u r a t , ± m h y z b .

Property Testing vs. “Statistics” Classical Statistics (Mean, Median, Quantiles): Also run in sublinear time. So what’s special about “linearity testing”? Classical statistics work on “symmetric” properties: Linear functions closed under much smaller group of permutations: Simlarly, graph properties have nice invariances F c l o s e d u n r a b i t y p m D . F c l o s e d u n r i a m p L : 2 ! j D l o g s u c h m a p v . !

Is testing a corollary of invariance? Example 1: Invariant group trivial. Testing easy. Example 2: Invariant group still trivial. No O(1) local tests. Conclusion: Testing not necessarily a consequence of invariance. If we believe this to be the case for the linearity test, must prove it!! D = R ; F 1 f I d j ( x ) g D = f 1 ; : n g R 9 F 2 j x < y ) (

Is testing a corollary of invariance? Example 1: Invariant group trivial. Testing easy. Example 2: Invariant group still trivial. No O(1) local tests. Conclusion: Testing not necessarily a consequence of invariance. If we believe this to be the case for the linearity test, must prove it!! D = R ; F 1 f I d j ( x ) g D = f 1 ; : n g R 9 F 2 j x < y ) (

Part II: Formal Definitions & Results

Linear Invariance Examples: F i s L n e a r I v t f ² l u b p c ( o ) j ) 2 d : ! ± ( A ± n e I v a r i c d ¯ s m l y ) ² L i n e a r f u c t o s , - v p l y m d g · h F 1 + 2

Testing, constraints, characterizations ² S u p o s e F h a k - q r y t . ² T h e n m b r s o f F a t i y k - l c . C o n s t r a i : = ( x 1 ; k 2 F u b p c e V µ ) 8 f ¯ . , h ² E . g , i n t h e l a r c s : f ( ® ) + ¯ = C = ( ® ; ¯ + V f 1 g ) C h a r c t e i z o n : = f 1 ; m g 2 F , s ¯

(Linear-Invariant) Algebraic Characterizations Characterizations require many constraints! Linear (affine) invariance turns one constraint into many. (Linearity) Example: C = ( x 1 ; : k V ) c o n s t r a i d L l e ± . ( L ® ) ; ¯ + V c o n s t r a i f l e y A l g e b r a i c C h t z o n : S s . f ± L j ( ) F

Main Theorems T h e o r m : F a ± n - i v t d s k l c g b z , p q u y . ² U n i ¯ e s , m p l a d x t r v o u g b c T h e o r m : F a ± n - i v t d s k l c ) f ( g b z .

X Pictorially F o r a ± n e - i v t p s k - q u e r y t s k ! k - l o Theorem 1 By defn. k ! k - l o c a h r t e i z n k - l o c a g . h r By defn. Theorem 2 k ! f F ( ) X k - l o c a n s t r i k ! (with Grigorescu)

X Pictorially F o r a ± n e - i v t p s k - q u e r y t s k ! k - l o Theorem 1 By defn. k ! k - l o c a h r t e i z n k - l o c a g . h r By defn. Theorem 2 k ! f F ( ) X k - l o c a n s t r i k ! (with Grigorescu)

Part III: BLR (and our) analysis

Step 1: Prove “most likely” is overwhelming majority. BLR Analysis: Outline ² H a v e f s . t P r x ; y [ ( ) + 6 = ] ± < 2 9 W n o h w c l m g F ² D e ¯ n g ( x ) = m o s t l i k y f + ¡ . ² I f c l o s e t F h n g w i b a d . ² B u t i f n o c l s e ? g m a y v b q d ¯ ! ² S t e p s : ¡ S t e p : P r o v f c l s g Step 1: Prove “most likely” is overwhelming majority. ¡ ¡ S t e p 2 : P r o v h a g i s n F .

BLR Analysis: Step 0 ² D e ¯ n g ( x ) = m o s t l i k y f + ¡ . C l a [ f ( ) 6 = g ] · 2 ± ¡ L e t B = f x j P r y [ ( ) 6 + ] ¸ 1 2 g ¡ P r x ; y [ l i n e a t s j c 2 B ] ¸ 1 ) P r x [ 2 B ] · ± ¡ I f x 6 2 B t h e n ( ) = g

BLR Analysis: Step 1 V o t e ( y ) ² D e ¯ n g ( x ) = m o s t l i k y f + ¡ . ² S u p o s e f r m x , 9 t w q a l y i k v . P b n d h c ? ² I f w e i s h t o g l n a r , d u c . L e m a : 8 x , P r y ; z [ V o t ( ) 6 = ] · 4 ±

BLR Analysis: Step 1 V o t e ( y ) ² D e ¯ n g ( x ) = m o s t l i k y f + ¡ . ² S u p o s e f r m x , 9 t w q a l y i k v . P b n d h c ? ² I f w e i s h t o g l n a r , d u c . L e m a : 8 x , P r y ; z [ V o t ( ) 6 = ] · 4 ±

BLR Analysis: Step 1 V o t e ( y ) ² D e ¯ n g ( x ) = m o s t l i k y f + ¡ . L e m a : 8 x , P r y ; z [ V o t ( ) 6 = ] · 4 ± f ( y ) ¡ f ( x + y ) ? f ( z ) f ( y + z ) ¡ f ( y + 2 z ) ¡ f ( x + z ) ¡ f ( 2 y + z ) f ( x + 2 y z ) P r o b . R w / c l u m n s - z e · ±

BLR Analysis: Step 1 V o t e ( y ) ² D e ¯ n g ( x ) = m o s t l i k y f + ¡ . L e m a : 8 x , P r y ; z [ V o t ( ) 6 = ] · 4 ± f ( y ) ¡ f ( x + y ) ? f ( z ) f ( y + z ) ¡ f ( y + 2 z ) ¡ f ( x + z ) ¡ f ( 2 y + z ) f ( x + 2 y z ) P r o b . R w / c l u m n s - z e · ±

BLR Analysis: Step 2 (Similar) f ± < 1 2 , t h n 8 x ; y g ( ) + = P r o b . R w / c l u m n s - z e · 4 ± g ( x ) g ( y ) ¡ g ( x + y ) f ( z ) f ( y + z ) ¡ f ( y + 2 z ) ¡ f ( x + z ) ¡ f ( 2 y + z ) f ( x + 2 y z )

Step 1: Prove “most likely” is overwhelming majority. Our Analysis: Outline ² f s . t P r L [ h ( x 1 ) ; : k i 2 V ] = ± ¿ ² D e ¯ n g ( x ) = ® t h a m i z s P r f L j 1 [ ; 2 : k V ] ² S t e p s : ¡ S t e p : P r o v f c l s g Step 1: Prove “most likely” is overwhelming majority. ¡ ¡ S t e p 2 : P r o v h a g i s n F .

Step 1: Prove “most likely” is overwhelming majority. Our Analysis: Outline ² f s . t P r L [ h ( x 1 ) ; : k i 2 V ] = ± ¿ ² D e ¯ n g ( x ) = ® t h a m i z s P r f L j 1 [ ; 2 : k V ] S a m e s b f o r ² S t e p s : ¡ S t e p : P r o v f c l s g Step 1: Prove “most likely” is overwhelming majority. ¡ ¡ S t e p 2 : P r o v h a g i s n F .

? Matrix Magic? V o t e ( L ) ² D e ¯ n g ( x ) = ® t h a m i z s P r f L j 1 [ ; 2 : k V ] L e m a : 8 x , P r ; K [ V o t ( ) 6 = ] · 2 k ¡ 1 ± L ( x 2 ) L ( x k ) ¢ x K ( x 2 ) ? . K ( x k )

? Matrix Magic? L ( x ) ¢ L ( x ) x K ( x ) . K ( x ) ² W a n t m r k 2 ) ¢ L ( x k ) x K ( x 2 ) ? . K ( x k ) ² W a n t m r k e d o w s b c i . ² S u p o s e x 1 ; : ` l i n a r y d t h m .

Matrix Magic? ` L ( x ) ¢ L ( x ) x K ( x ) ` . K ( x ) ² S u p o s e Fill with random entries Matrix Magic? Fill so as to make constraints Linear algebra implies final columns are also constraints. ` L ( x 2 ) ¢ L ( x k ) x K ( x 2 ) ` . K ( x k ) ² S u p o s e x 1 ; : ` l i n a r y d t h m .

Matrix Magic? ` L ( x ) ¢ L ( x ) x K ( x ) ` . K ( x ) ² S u p o s e Fill with random entries Matrix Magic? Fill so as to make constraints Linear algebra implies final columns are also constraints. ` L ( x 2 ) ¢ L ( x k ) x K ( x 2 ) ` . K ( x k ) ² S u p o s e x 1 ; : ` l i n a r y d t h m .

Conclusions Linear/Affine-invariant properties testable if they have local constraints. Gives clean generalization of linearity and low-degree tests. Future work: What kind of invariances lead to testability (from characterizations)?