1 / 13

Computational Analogues of Entropy

Computational Analogues of Entropy. Boaz Barak Ronen Shaltiel Avi Wigderson. Our Objectives:. 1. Investigate possible defs for computational Min-Entropy. 2. Check whether computational defs satisfy analogs of statistical properties. Statistical Min-Entropy.

eunice
Télécharger la présentation

Computational Analogues of Entropy

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Computational Analogues of Entropy Boaz BarakRonen ShaltielAvi Wigderson

  2. Our Objectives: 1. Investigate possible defs forcomputationalMin-Entropy. 2. Check whether computational defs satisfy analogs of statistical properties. Statistical Min-Entropy Definition:H(X)¸k iff maxx Pr[ X=x ]<2-k ( X r.v. over {0,1}n ) Properties: • H(X) · Shannon-Ent(X) • H(X)=n iff X~Un • H(X,Y) ¸ H(X) (concatenation) • If H(X)¸k then 9(efficient)fs.t.f(X)~Uk/2(extraction)

  3. Our Contributions • Study 3 variants (1 new) of pseudoentropy. • Equivalence & separation results for several computational model. • Study analogues of IT results. In this talk: • Present the 3 variants. • Show 2 results + proof sketches

  4. Review - Pseudorandomness Def:X is pseudorandom if maxD2C biasD(X,Un) <  i.e., X is computationally indistinguishable from Un C– class of efficient algorithms (e.g. s-sized circuits) biasD(X,Y) =| EX[D(X)] - EY[D(Y)] |  – parameter (in this talk: some constant > 0)

  5. i.e., X is computationally indist. from someY with ¸k statistical min-entropy. i.e., 8 efficient D, X is computationally indist. by D from someY=Y(D) with ¸k statistical min-entropy. Defining Pseudoentropy Def 1 [HILL]: HHILL(X)¸k if 9Y s.t. H(Y)¸ k and maxD2C biasD(X,Y) <  minH(Y)¸ K maxD2C biasD(X,Y) <  Def 2: HMet(X)¸k if maxD2C minH(Y)¸ K biasD(X,Y) <  Def 3 [Yao]: HYao(X)¸k if X cannot be efficiently compressed to k-1 bits. maxD2C biasD(X,Un) <  *X is pseudorandom if

  6. Defining Pseudoentropy HHILL(X)¸k if minH(Y)¸ K maxD2C biasD(X,Y) <  HMet(X)¸k if maxD2C minH(Y)¸ K biasD(X,Y) <  HYao(X)¸k if X can’t be efficiently compressed tok-1bits. Claim 1: H(X) · HHILL(X) · HMet(X) · HYao(X) Claim 2: Fork=n all 3 defs equivalent to pseudorandomness. Claim 3: All 3 defs satisfy extraction property.[Tre]

  7. 2: Use the “Min-Max” theorem. [vN28] HILL & Metric Def are Equivalent (For C = poly-sized circuits, any ) Thm 1: HHILL(X) = HMet(X) Proof: SupposeHHILL(X)<k Player 2: D Player 1:D Y Y Player 1: biasD(X,Y)¸ HHILL(X)¸k if minH(Y)¸K maxD2C biasD(X,Y) <  HMet(X)¸k if maxD2C minH(Y)¸K biasD(X,Y) < 

  8. Can we do better? Unpredictability & Entropy Thm [Yao]: If X is unpredicatble with adv.  then X is pseudorandom w/ param ’=n¢ Loss of factor of n due to hybrid argument –useless for constant advantage  This loss can be crucial for some applications (e.g., extractors, derandomizing small-space algs)

  9. Unpredictability & Entropy IT Fact [TZS]: If X is IT-unpredictable with const. adv. then H(X)=(n) We obtain the following imperfect analog: Thm 2: If X is unpredictable by SAT-gate circuits with const. adv. then HMet(X)=(n) In paper: A variant of Thm 2 for nonuniform online logspace.

  10. {0,1}n D X Thm 2: If X is unpredictable by SAT-gate circuits with const. adv. then HMet(X)=(n) Proof: Suppose thatHMet(X)<n We’ll construct a SAT-gate predictor P s.t. Pri,X[ P(X1,…,Xi-1)=Xi] = 1 –  We have that maxD2CminH(Y)¸n biasD(X,Y)¸ i.e., 9D s.t.8Y If H(Y)¸n then biasD(X,Y)¸ Assume: 1) |D-1(1)| < 2n*2) PrX[ D(X)=1 ] = 1

  11. {0,1}n D X Construct P from D 1) |D-1(1)| < 2n2) PrX[ D(X)=1 ] = 1 Define predictor P as follows:P(x1,…,xi)=0 iff Pr[ D(x1,…,xi,0,Un-i-1)=1] > ½ Note that P does not depend on X and can be constructed w/ NP oracle. (approx counting [JVV]) Claim: 8x2D, Ppredicts at least (1-)n indices ofx

  12. Claim: 8x2D, Ppredicts at least(1-)n indices of x Proof: SupposePfails to predictx in m indices. ¸2m ¸8 We’ll show that |D|>2m,obtaining a contradiction. ¸4 ¸4 ¸2 ¸2 1 P(x1,…,xi)=0 iff Pr[ D(x1,…,xi,0,Un-i-1)=1] > ½

  13. Open Problems More results for poly-time computation: • Analog of Thm 2 (unpredictabilityentropy)? • Meaningful concatenation property? • Separate Yao & Metric pseudoentropy. Prove that RL=L

More Related