closure amortization lower bounds and
play

Closure, Amortization, Lower-bounds, and Separations Benny - PowerPoint PPT Presentation

Conditional Disclosure of Secrets: Amplification, Closure, Amortization, Lower-bounds, and Separations Benny Applebaum Barak Arkis Pavel Raykov Prashant Nalini Vasudevan Conditional Disclosure of Secrets [GIKM00] : 0,1 0,1


  1. Conditional Disclosure of Secrets: Amplification, Closure, Amortization, Lower-bounds, and Separations Benny Applebaum Barak Arkis Pavel Raykov Prashant Nalini Vasudevan

  2. Conditional Disclosure of Secrets [GIKM00] 𝑔: 0,1 π‘œ Γ— 0,1 π‘œ β†’ {0,1} 𝜺 -Correctness: If 𝑔 𝑦, 𝑧 = 1, then for any 𝑑 , Randomness 𝑠 Pr 𝐷 𝑦, 𝑧, 𝑛 𝐡 , 𝑛 𝐢 = 𝑑 > 1 βˆ’ Ξ΄ 𝑦 𝑧 A B Secret 𝑑 𝝑 -Privacy: If 𝑔 𝑦, 𝑧 = 0 , then for any 𝑑 , 𝑛 𝐡 𝑛 𝐢 Ξ” 𝑇𝑗𝑛 𝑦, 𝑧 ; 𝑛 𝐡 , 𝑛 𝐢 < πœ— Communication: 𝑛 𝐡 + |𝑛 𝐢 | C 𝑦, 𝑧 Randomness: |𝑠|

  3. Connections and Applications β€’ Attribute-Based Encryption. [Att14,Wee14] β€’ Secret-sharing for certain graph-based access structures. β€’ Light-weight alternative to zero-knowledge proofs in some settings. [AIR01] β€’ Data privacy in information-theoretic PIR. [GIKM00] β€’ A minimal model of multi-party computation.

  4. What Was Known Earlier Upper bounds: β€’ Communication 2 𝑃( π‘œ log π‘œ) for any predicate on π‘œ -bit inputs. [LVW17] β€’ Communication 𝑃(𝜏) for predicates with size- 𝜏 branching programs or span programs. [IW14,AR16] Lower bounds: β€’ Explicit predicate that requires Ξ©(log π‘œ) bits of communication. [GKW15] β€’ Same predicate requires Ξ© π‘œ bits for linear CDS. [GKW15]

  5. CDS and Statistical Difference Randomness 𝑠 𝜺 -Correctness: 𝑧 𝑦 A B If 𝑔 𝑦, 𝑧 = 1, then for any 𝑑 , Secret 𝑑 Pr 𝐷 𝑦, 𝑧, 𝑛 𝐡 , 𝑛 𝐢 = 𝑑 > 1 βˆ’ Ξ΄ 𝑛 𝐡 𝑛 𝐢 0 1 ≑ Ξ” 𝑛 𝐡 , 𝑛 𝐢 𝑦,𝑧 ; 𝑛 𝐡 , 𝑛 𝐢 𝑦,𝑧 > 1 βˆ’ 2πœ€ C 𝑦, 𝑧 𝝑 -Privacy: If 𝑔 𝑦, 𝑧 = 0 , then for any 𝑑 , Distribution of (𝑛 𝐡 , 𝑛 𝐢 ) : Ξ” 𝑇𝑗𝑛 𝑦, 𝑧 ; 𝑛 𝐡 , 𝑛 𝐢 < πœ— 0 β€’ input (𝑦, 𝑧) , 𝑑 = 0 : 𝑛 𝐡 , 𝑛 𝐢 𝑦,𝑧 β€’ input (𝑦, 𝑧) , 𝑑 = 1 : 𝑛 𝐡 , 𝑛 𝐢 𝑦,𝑧 1 0 1 ≑ Ξ” 𝑛 𝐡 , 𝑛 𝐢 𝑦,𝑧 ; 𝑛 𝐡 , 𝑛 𝐢 𝑦,𝑧 < 2πœ—

  6. Separations Explicit function π‘„π·π‘π‘š: 0,1 4n log π‘œ Γ— 0,1 2n log π‘œ β†’ 0,1 that has: β€’ CDS complexity: 𝑃(log π‘œ) β€’ Randomized communication complexity: Ξ©(π‘œ 1/3 ) β€’ Linear CDS complexity: Ξ©(π‘œ 1/6 ) Inspired by oracle separations between SZK and other classes [Aar12], and the Pattern Matrix method [She11].

  7. Collision Problems π‘œ log π‘œ β„Ž 𝑨 : 0,1 log π‘œ β†’ 0,1 log π‘œ 𝑨 β‹… β‹… β‹… β„Ž 𝑨 𝑗 = 𝑗 π‘’β„Ž block in 𝑨 log π‘œ π‘œ blocks β‡’ β„Ž 𝑨 (𝑗) is uniformly distributed 0 if β„Ž 𝑨 is 1βˆ’toβˆ’1 π·π‘π‘š 𝑨 = ቐ 1 if β„Ž 𝑨 is 2βˆ’toβˆ’1 β‡’ β„Ž 𝑨 (𝑗) is far from uniform

  8. Collision Problems 4π‘œ log π‘œ 𝑦 β‹… β‹… β‹… 0 0 1 0 1 0 1 0 0 1 1 1 1 0 0 1 𝑧 π‘„π·π‘π‘š 𝑦, 𝑧 = π·π‘π‘š(𝑦 𝑧 ) 3 2 2 β‹… β‹… β‹… 4 𝑦[𝑧] 1 0 1 β‹… β‹… β‹… 1 π‘œ log π‘œ 𝑆 π‘„π·π‘π‘š > Ξ©(π‘œ 1/3 ) linCDS π‘„π·π‘π‘š > Ξ©(π‘œ 1/6 ) ([Amb05,Kut05] + [She11]) (left + [GKW15])

  9. Collision Problems 𝑦 A 0 0 1 0 1 0 1 0 0 1 1 1 β‹… β‹… β‹… 1 0 0 1 𝑦, 𝑧 𝑑 𝑗 C 𝑧 B 3 2 2 β‹… β‹… β‹… 4 Use PSM [FKN94] to send: β€’ β„Ž 𝑦 𝑧 (𝑗) if 𝑑 = 0 𝑦[𝑧] β€’ 𝑠 ← 0,1 log π‘œ if 𝑑 = 1 1 0 1 β‹… β‹… β‹… 1 log π‘œ If π‘„π·π‘π‘š 𝑦, 𝑧 = 0 , both are π‘œ blocks the same distribution, else they are far apart.

  10. Closure β„Ž - Boolean formula over 0,1 𝑛 of size 𝜏 CDS for each of CDS for 𝑔 1 , … , 𝑔 β„Ž(𝑔 1 , … , 𝑔 𝑛 ) 𝑛 Comm: 𝑒 1 , … , 𝑒 𝑛 Comm: 𝜏 β‹… π‘žπ‘π‘šπ‘§(𝑒 𝑗 , 𝜍 𝑗 ) Rand : 𝜍 1 , … , 𝜍 𝑛 Rand : 𝜏 β‹… π‘žπ‘π‘šπ‘§(𝑒 𝑗 , 𝜍 𝑗 ) Construction uses transformations for Statistical Difference [SV03,Oka96], and PSM protocols [FKN94].

  11. Amplification CDS for 𝑔 CDS for 𝑔 Single-bit secret 𝑙 -bit secret Corr: 2 βˆ’Ξ©(𝑙) Corr: 0.1 Priv: 2 βˆ’Ξ©(𝑙) Priv: 0.1 Comm: 𝑒 Comm: 𝑃(𝑙𝑒) Construction uses constant-rate ramp secret-sharing schemes [CCGdHV07]. Incomparable version follows from the Polarization Lemma [SV03].

  12. Lower Bound There exists a predicate 𝑔: 0,1 π‘œ Γ— 0,1 π‘œ β†’ {0,1} for which any perfect (single-bit) CDS requires communication at least 0.99π‘œ . Proven by reduction to the PSM lower bound of [FKN94]. Earlier bound was explicit, Ξ©(log π‘œ) bits. [GKW15]

  13. Amortization For any predicate 𝑔: 0,1 π‘œ Γ— 0,1 π‘œ β†’ {0,1} and 𝑛 > 2 2 2π‘œ , there is a perfect CDS protocol for 𝑔 with 𝑛 -bit secrets with communication complexity 𝑃(π‘›π‘œ) . Proven using techniques from the amortization of branching programs [Pot16]. 𝑛 -fold repetition of best known general protocol [LVW17]: 𝑛 β‹… 2 𝑃( π‘œ log π‘œ)

  14. Summary We prove the following properties of CDS: β€’ Lower Bounds: Non-explicit, Ξ©(π‘œ) . β€’ Separation: From insecure communication and linear CDS. β€’ Amortization: 𝑃(π‘œ) per bit of secret, if there are more than 2 2 2π‘œ bits. β€’ Closure: Under composition with formulas. β€’ Amplification: Of correctness and privacy from constant to 2 βˆ’Ξ©(𝑙) with 𝑃(𝑙) blowup. To note: β€’ Connections with Statistical Difference and SZK. β€’ Barriers to PSM lower bounds.

Recommend


More recommend