Download presentation
Presentation is loading. Please wait.
Published byBeverley Jacobs Modified over 9 years ago
1
The Closest Vector is Hard to Approximate and now, for unlimited time only with Pre - Processing !! Nisheeth vishnoi Subhash Khot Michael Alekhnovich Joint work with Guy Kindler Microsoft Research
2
In this talk: In this talk: Lattices Lattices The closest vector problem: background The closest vector problem: background Our results: NP-hardness for CV-PP Our results: NP-hardness for CV-PP Proving hardness with preprocessing Proving hardness with preprocessing Something about our proof: new property of PCPs Something about our proof: new property of PCPs
3
A lattice, L: A discrete additive subgroup of R n. A basis for L: b 1,…,b n 2 R n, s.t. L={ i a i b i : a 1,..,a n 2 Z }.
4
The Closest Vector Problem ( CVP )
5
CVP : Given a lattice L and a target vector t, find the point in L closest to t in l p distance. [Regev Ronen 05] Hardness results in l 2 carry for any l p. [Ajtai Kumar Sivakumar 01]: 2 O(nloglog(n)/log n) =2 o(n) approx. [Dinur Kindler Raz Safra 98]: n O(1/loglog n) =n o(1) hardness. [Lagarias Lenstra Schnorr 90, Banaszczyk 93, Goldreich Goldwasser 00, Aharonov Regev 04] NP-hardness of (n/log n) 1/2 would collapse the polynomial hierarchy.
6
Motivation for studying CVP [Ajtai 96]: Worst case to average case reductions for lattice problems. [Ajtai Dwork 97] Based cryptosystems on lattice problems. [Goldreich Goldwasser Halevi 97] Cryptosystem based on CVP. [Micciancio Vadhan 03] Identification scheme based on (n/log n) 1/2 hardness for CVP. t L t – message. L – coding function: known in advance, and reused.
7
Is it safe to reuse L as key? CV-PP : Preprocess L for unlimited time, Given t, solve CVP on L,t. [Kannan 87, Lagarias Lenstra Schnorr 90, Aharonov Regev ] O(n 1/2 )-approx. for CV-PP. [Feige Micciancio 02] (5/3) 1/p approx. hardness for CV-PP. [Regev 03] 3 1/p approx. hardness for CV-PP.
8
Our Results Thm: CV-PP in NP-hard(!) to approximate within any constant. Also applies to NC-PP. Unless NP µ DTIME(2 polylog n ), NC-PP is hard to approximate within (log n) 1- NC-PP is hard to approximate within (log n) 1- CV-PP is hard to approximate within (log n) (1/p)- CV-PP is hard to approximate within (log n) (1/p)- 1st Proof : By reduction from E-k-HVC [DGKR 03]. 2nd proof: Using PCP-PP constructions, plus smoothing technique of [Khot 02].
9
Proving hardness with preprocessing Hardness of approximation within gap g: I 2 ¦ ) dist(t,L) · d I ¦ ) dist(t,L) ¸ d ¢ g I : Instance of ¦ 2 NPC Reduction L, t
10
Proving hardness with preprocessing I : Instance of ¦ 2 NPC Reduction L, t Hardness of approximation within g, with preprocessing: Size of I Partial Input Generator Preprocessed L CV-PP t t I 2 ¦ ) dist(t,L) · d I ¦ ) dist(t,L) ¸ d ¢ g Hardness of approximation within gap g:
11
Size of I Partial Input Generator I : Instance of ¦ 2 NPC Reduction PCP with preprocessing ( PCP-PP ) Preprocessed L t t CV-PP LEFT RIGHT PCP-PP I 2 ¦ ) dist(t,L) · d I ¦ ) dist(t,L) ¸ d ¢ g PCP : Gap version of Q uadratic equations. x 2 +2xy=7 x 2 +z 2 =5..
12
Size of I Partial Input Generator I : Instance of ¦ 2 NPC Reduction PCP with preprocessing ( PCP-PP ) LEFT RIGHT PCP-PP I 2 ¦ ) opt(LEFT,RIGHT)=1 I ¦ ) opt(LEFT,RIGHT) · c<1 PCP : Gap version of Q uadratic equations. x 2 +2xy=7 x 2 +z 2 =5..
13
Size of I Partial Input Generator I : Instance of ¦ 2 NPC Reduction PCP with preprocessing ( PCP-PP ) LEFT RIGHT PCP-PP PCP : Gap version of Q uadratic equations.
14
Size of I I : Instance of ¦ 2 NPC PCP with preprocessing ( PCP-PP ) LEFT RIGHT Preprocessed L t t CV-PP PCP : Gap version of Q uadratic equations. PCP-PP
15
PCP with preprocessing ( PCP-PP ) LEFT RIGHT PCP-PP PCP : Gap version of Q uadratic equations.
16
PCP-PP construction LEFT RIGHT PCP-PP PCP : Gap version of Q uadratic equations. Just (carefully) apply usual PCP construction!
17
Open problems Get better hardness parameters for CV-PP (perhaps using methods from [DKRS 98] ). Get improved hardness results for lattice problems, under stronger assumptions than NP P. Find more uses for PCP-PP constructions.
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.