np hardness of
play

NP-hardness of Minimum Circuit Size Problem for OR-AND-MOD Circuits - PowerPoint PPT Presentation

NP-hardness of Minimum Circuit Size Problem for OR-AND-MOD Circuits Shuichi Hirahara (The University of Tokyo) Igor C. Oliveira (University of Oxford) Rahul Santhanam (University of Oxford) CCC 2018 @ San Diego June 22, 2018 Talk Outline


  1. NP-hardness of Minimum Circuit Size Problem for OR-AND-MOD Circuits Shuichi Hirahara (The University of Tokyo) Igor C. Oliveira (University of Oxford) Rahul Santhanam (University of Oxford) CCC 2018 @ San Diego June 22, 2018

  2. Talk Outline 1. MCSP and Its background 2. π’Ÿ - MCSP for a circuit class π’Ÿ 3. Our Results 4. Proof Sketch

  3. Talk Outline 1. MCSP and Its background 2. π’Ÿ - MCSP for a circuit class π’Ÿ 3. Our Results 4. Proof Sketch

  4. Minimum Circuit Size Problem (MCSP) Output Input β€’ Truth table π‘ˆ ∈ 0,1 2 𝑒 of a Is there a circuit of size ≀ 𝑑 function 𝑔: 0,1 𝑒 β†’ 0,1 that computes 𝑔 ? β€’ Size parameter 𝑑 ∈ β„• Example: Output: β€œYES” 𝑑 = 5 π’š 𝟐 π’š πŸ‘ π’š 𝟐 βŠ• π’š πŸ‘ 0 0 0 0 1 1 𝑔 = 1 0 1 1 1 0

  5. Brief History of MCSP 1950s Recognized as an important problem in the Soviet Union [Trakhtenbrot ’s survey] 1970s Levin delayed publishing his work because he wanted to say something about MCSP. Masek proved NP-completeness of DNF -MCSP. 1979 2000 Kabanets and Cai revived interest, based on natural proofs. [Razborov & Rudich (1997)] Since then many papers and results appeared; however, the complexity of MCSP remains elusive.

  6. Current Knowledge about MCSP ➒ Upper bound: MCSP ∈ 𝐎𝐐 ➒ Lower bound: βˆƒ pseudorandom function generators ⟹ MCSP βˆ‰ 𝐐 ➒ Big Open Question: Is MCSP NP-hard? ➒ No consensus about the exact complexity of MCSP βœ“ No strong evidence against NP-completeness β€’ Weak evidence: [Hirahara- Santhanam (CCC’17)] [Allender - Hirahara 17]… βœ“ No strong evidence for NP-completeness β€’ Some new evidence: [Impagliazzo-Kabanets-Volkvovich (CCC’18)] & This work

  7. Kabanets-Cai Obstacle: Why so difficult? ➒ Suppose that we want to construct a reduction from SAT to MCSP . ↦ (𝑔, 𝑑) πœ’ ∈ SAT CircSize 𝑔 ≀ 𝑑 ↦ (𝑔, 𝑑) CircSize 𝑔 > 𝑑 πœ’ βˆ‰ SAT Need to prove a circuit lower bound! ➒ Natural reduction techniques would imply E ⊈ SIZE π‘œ 𝑃 1 . [Kabanets-Cai (2000)]

  8. Talk Outline 1. MCSP and Its background 2. π’Ÿ - MCSP for a circuit class π’Ÿ 3. Our Results 4. Proof Sketch

  9. π’Ÿ - MCSP for a circuit class π’Ÿ Output Input β€’ Truth table π‘ˆ ∈ 0,1 2 𝑒 of a Is there a π’Ÿ -circuit of size ≀ 𝑑 function 𝑔: 0,1 𝑒 β†’ 0,1 that computes 𝑔 ? β€’ Size parameter 𝑑 ∈ β„• Theorem [Masek (1978 or 79, unpublished)] DNFβˆ’MCSP is NP-hard.

  10. DNF - MCSP Output Input β€’ Truth table π‘ˆ ∈ 0,1 2 𝑒 of a Is there a DNF formula of size function 𝑔: 0,1 𝑒 β†’ 0,1 ≀ 𝑑 that computes 𝑔 ? β€’ Size parameter 𝑑 ∈ β„• Depth: 2 ∨ Size: 3 Example of DNFs: ∧ ∧ ∧ ≑ ¬𝑦 1 ∧ 𝑦 2 ∨ 𝑦 2 ∧ ¬𝑦 3 ∨ ¬𝑦 2 ¬𝑦 1 𝑦 2 𝑦 2 ¬𝑦 3 ¬𝑦 2 The size of DNF ≔ #(clauses)

  11. π’Ÿ - MCSP for π’Ÿ βŠƒ DNF ➒ Beyond DNFs , no NP -hardness was proved since the work of Masek (1979). ➒ To quote Allender, Hellerstein, McCabe, Pitassi, and Saks (2008): β€œThus an important open question is to resolve the NP- hardness of … function minimization results above for classes that are stronger than DNF.”

  12. Known results about π’Ÿ - MCSP More expressive MCSP Hardness based on cryptography β‹― e.g.) Blum integer 0 - MCSP for large 𝑒 factorization [AHMPS08] AC 𝑒 β‹― No hardness result Depth3- MCSP Known to be NP-hard DNF - MCSP [Masek (1979)] Remark: The complexity is not necessarily monotone increasing or decreasing.

  13. Talk Outline 1. MCSP and Its background 2. π’Ÿ - MCSP for a circuit class π’Ÿ 3. Our Results 4. Proof Sketch

  14. Our Results ➒ The first NP-hardness result for π’Ÿ -MCSP for a class π’Ÿ βŠƒ DNF Theorem (Main Result) DNF ∘ XOR βˆ’MCSP is NP-hard under polynomial-time many-one reductions. ➒ Our proof techniques extend to: β€’ DNF ∘ MOD 𝑛 - MCSPβ€² is NP-hard for any 𝑛 β‰₯ 2 , but the input is a truth table of an 𝑛 -valued function 𝑔: β„€/𝑛℀ 𝑒 β†’ 0,1 .

  15. DNF ∘ XOR circuits ( 2 Ξ© π‘œ circuit lower bound is known) [Cohen & Shinkar (2016)] ∨ Example Depth 3 ∧ ∧ ∧ Size 3 1 st layer: an OR gate 2 nd layer: AND gates βŠ• βŠ• βŠ• βŠ• 3 rd layer: XOR gates 𝑦 2 ¬𝑦 1 ¬𝑦 3 ¬𝑦 2 The size of DNF ∘ XOR circuits ≔ The number of AND gates ➒ This is a convenient circuit size measure as advocated by Cohen & Shinkar (2016). 1. Nice combinatorial meaning 2. W.l.o.g., # XOR gates ≀ π‘œ β‹… #(AND gates) ➒ Our proof techniques extend to the number of all the gates in a DNF ∘ XOR formula.

  16. DNF ∘ XOR circuits ( 2 Ξ© π‘œ circuit lower bound is known) [Cohen & Shinkar (2016)] ∨ Example Depth 3 ∧ ∧ ∧ Size 3 1 st layer: an OR gate 2 nd layer: AND gates βŠ• βŠ• βŠ• βŠ• 3 rd layer: XOR gates 𝑦 2 ¬𝑦 1 ¬𝑦 3 ¬𝑦 2 1 βŠ• 𝑦 1 βŠ• 𝑦 2 βŠ• 1 βŠ• 𝑦 3 = 1 outputs 1. ⟺ The subcircuit 𝑦 2 βŠ• 1 βŠ• 𝑦 3 = 1 ← Some linear equations over GF 2 ⟺ 𝑦 1 , 𝑦 2 , 𝑦 3 ∈ 𝐡 (for some affine subspace 𝐡 βŠ† GF 2 π‘œ )

  17. DNF ∘ XOR circuits ( 2 Ξ© π‘œ circuit lower bound is known) [Cohen & Shinkar (2016)] 𝑔: 0,1 π‘œ β†’ 0,1 ∨ Example Depth 3 𝐡 3 𝐡 1 𝐡 2 ∧ ∧ ∧ Size 3 1 st layer: an OR gate 2 nd layer: AND gates βŠ• βŠ• βŠ• βŠ• 3 rd layer: XOR gates 𝑦 2 ¬𝑦 1 ¬𝑦 3 ¬𝑦 2 𝑔 βˆ’1 1 = 𝐡 1 βˆͺ 𝐡 2 βˆͺ 𝐡 3

  18. The Important Observation The minimum DNF ∘ XOR circuit size for computing 𝑔 = The minimum number 𝑛 of affine subspaces needed to cover 𝑔 βˆ’1 1 : that is, βˆƒπ΅ 1 , … , 𝐡 𝑛 : affine subspaces of 0,1 π‘œ 𝐡 𝑗 βŠ† 𝑔 βˆ’1 1 𝐡 1 βˆͺ β‹― βˆͺ 𝐡 𝑛 = 𝑔 βˆ’1 1 and

  19. Talk Outline 1. MCSP and Its background 2. π’Ÿ - MCSP for a circuit class π’Ÿ 3. Our Results 4. Proof Sketch

  20. ➒ Our proof was inspired by a simple proof of Masek’s result given by [Allender, Hellerstein, McCabe, Pitassi, and Saks (2008)]. ➒ We extend and generalize their ideas significantly.

  21. Proof Outline Theorem (Main Result) π‘ž NP ≀ 𝑛 DNF ∘ XOR βˆ’MCSP Step 1. 2-factor approx. of DNF ∘ XOR - MCSP ZPP ≀ 𝑛 𝑠 -Bounded Set Cover for partial functions (NP-hard [Trevisan 2001] ) DNF ∘ XOR - MCSP Step 2. ZPP DNF ∘ XOR - MCSP ≀ 𝑛 for partial functions Derandomization using πœ— -biased generators Step 3. [Naor & Naor (1993)]

  22. Proof Outline Theorem (Main Result) π‘ž NP ≀ 𝑛 DNF ∘ XOR βˆ’MCSP Step 1. 2-factor approx. of DNF ∘ XOR - MCSP ZPP ≀ 𝑛 𝑠 -Bounded Set Cover for partial functions (NP-hard [Trevisan 2001] ) DNF ∘ XOR - MCSP Step 2. ZPP DNF ∘ XOR - MCSP ≀ 𝑛 for partial functions Derandomization using πœ— -biased generators Step 3. [Naor & Naor (1993)]

  23. The Set Cover Problem A universe 𝑉 and a collection of sets 𝒯 βŠ† 2 𝑉 Input: Output: The minimum π’Ÿ such that π’Ÿ βŠ† 𝒯 and Ϊ‚ π·βˆˆπ’Ÿ 𝐷 = 𝑉 Example: 𝑉 = { … } , 𝒯 = { … }

  24. The Set Cover Problem A universe 𝑉 and a collection of sets 𝒯 βŠ† 2 𝑉 Input: Output: The minimum π’Ÿ such that π’Ÿ βŠ† 𝒯 and Ϊ‚ π·βˆˆπ’Ÿ 𝐷 = 𝑉 Example: 𝑉 = { … } , 𝒯 = { … } A minimum cover π’Ÿ :

  25. The 𝑠 -Bounded Set Cover Problem A universe 𝑉 and a collection of sets 𝒯 βŠ† 2 𝑉 Input: such that 𝑇 ≀ 𝑠 for every 𝑇 ∈ 𝒯 . Output: The minimum π’Ÿ such that π’Ÿ βŠ† 𝒯 and Ϊ‚ π·βˆˆπ’Ÿ 𝐷 = 𝑉 Example: 𝑉 = { … } , 𝒯 = { … } 4-bounded, but not 3-bounded ➒ We set 𝑠 to be large enough [Feige (1998)] [Trevisan (2001)] so that a 2 -factor approx. is NP-hard. Approximation of 1 βˆ’ 𝑝 1 ln 𝑠 is NP -hard.

  26. Proof Outline Theorem (Main Result) π‘ž NP ≀ 𝑛 DNF ∘ XOR βˆ’MCSP Step 1. 2-factor approx. of DNF ∘ XOR - MCSP ZPP ≀ 𝑛 𝑠 -Bounded Set Cover for partial functions (NP-hard [Trevisan 2001] ) DNF ∘ XOR - MCSP Step 2. ZPP DNF ∘ XOR - MCSP ≀ 𝑛 for partial functions Derandomization using πœ— -biased generators Step 3. [Naor & Naor (1993)]

  27. DNF ∘ XOR - MCSP βˆ— for partial functions Output Input β€’ Truth table of a partial Is there a circuit of size ≀ 𝑑 function that agrees with 𝑔 on inputs 𝑔: 0,1 𝑒 β†’ 0,1,βˆ— from 𝑔 βˆ’1 0,1 ? β€’ Size parameter 𝑑 ∈ β„• Example: π’š 𝟐 π’š πŸ‘ π’ˆ π’š 𝟐 , π’š πŸ‘ 0 0 βˆ— 0 1 1 𝑔 = βˆ— 1 0 1 1 0

  28. Claim DNF ∘ XOR - MCSP 2-factor approx. of ZPP ≀ 𝑛 for partial functions 𝑠 -Bounded Set Cover ➒ Given: 𝑉 = 1, … , 𝑂 , 𝒯 = {𝑇 1 , … , 𝑇 𝑛 } Construct 𝑔: 0,1 𝑒 β†’ 0,1,βˆ— for 𝑒 = 𝑃 log 𝑂 ➒ Goal: DNF ∘ XOR - MCSP Set Cover 𝑀 𝑗 ∼ 0,1 𝑒 (A uniformly random vector) 𝑗 ∈ 𝑉 ↦ span π‘—βˆˆπ‘‡ π‘˜ 𝑀 𝑗 βŠ† 0,1 𝑒 𝑇 π‘˜ ∈ 𝒯 ↦ span π‘—βˆˆπ‘‡ 𝑀 𝑗 βŠ† 0,1 𝑒 Cover π’Ÿ βŠ† 𝒯 ↦ ራ π‘‡βˆˆπ’Ÿ

  29. Set Cover DNF ∘ XOR - MCSP 𝑔: 0,1 𝑒 β†’ 0,1,βˆ— 𝑉 = 1,2,3 , 𝒯 = {𝑇 1 , 𝑇 2 } 𝑇 1 = 1,2 1 2 3 𝑇 2 = 2,3 𝑔 𝑀 𝑗 ≔ 1 for any 𝑗 ∈ 𝑉 . ➒ 𝑔 𝑦 ≔ 0 for all 𝑦 βˆ‰ span 𝑀 1 , 𝑀 2 βˆͺ span 𝑀 2 , 𝑀 3 . ➒ ➒ 𝑔 𝑧 ≔ βˆ— for any other vector 𝑧 ∈ 0,1 𝑒 . β€’ The minimum DNF ∘ XOR circuit size for computing 𝑔 The minimum number of affine subspaces 𝐡 βŠ† 𝑔 βˆ’1 1,βˆ— = needed to cover 𝑔 βˆ’1 1 = 𝑀 1 , 𝑀 2 , 𝑀 3 .

Recommend


More recommend