blaze blazing fast privacy preserving machine learning
play

BLAZE: BLAZING FAST PRIVACY-PRESERVING MACHINE LEARNING ARPITA PATRA - PowerPoint PPT Presentation

BLAZE: BLAZING FAST PRIVACY-PRESERVING MACHINE LEARNING ARPITA PATRA AND AJITH SURESH Ajith Suresh CrIS Lab, IISc https://www.csa.iisc.ac.in/~cris Outline q Secure Multi-party Computation (MPC) q MPC for small number of parties (3PC) q Our


  1. BLAZE: BLAZING FAST PRIVACY-PRESERVING MACHINE LEARNING ARPITA PATRA AND AJITH SURESH Ajith Suresh CrIS Lab, IISc https://www.csa.iisc.ac.in/~cris

  2. Outline q Secure Multi-party Computation (MPC) q MPC for small number of parties (3PC) q Our Efficient BLAZE Protocol (Results) q Privacy Preserving Machine Learning (PPML) AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  3. Secure Multi-party Computation (MPC) [Yao’82] ü A set of parties with private inputs wish to compute some joint function of their inputs. ü Goals of MPC: Correctness – Parties should correctly evaluate the function § output. Privacy – Nothing more than the function output should be § revealed AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  4. Secure Multi-party Computation (MPC) [Yao’82] ü A set of parties with private inputs wish to compute some joint function of their inputs. ü Goals of MPC: Correctness – Parties should correctly evaluate the function § Trusted output. Third Party Privacy – Nothing more than the function output should be § revealed AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  5. Secure Multi-party Computation (MPC) [Yao’82] ü A set of parties with private inputs wish to compute some joint function of their inputs. ü Goals of MPC: Correctness – Parties should correctly evaluate the function § Trusted output. Third Party Privacy – Nothing more than the function output should be § revealed AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  6. Secure Multi-party Computation (MPC) [Yao’82] ü A set of parties with private inputs wish to compute some joint function of their inputs. ü Goals of MPC: Correctness – Parties should correctly evaluate the function § MPC output. Privacy – Nothing more than the function output should be § revealed AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  7. Secure Multi-party Computation (MPC) [Yao’82] ADVERSARY • Semi – honest: • Follows the protocol but tries to learn more MPC • Malicious: • Can arbitrarily deviate from the protocol AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  8. Secure Multi-party Computation (MPC) [Yao’82] ADVERSARY • Semi – honest: • Follows the protocol but tries to learn more MPC • Malicious: • Can arbitrarily deviate from the protocol Malicious Corruption AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  9. MPC for small number of parties Ø Efficiency and Simplicity [MRZ15,AFLNO16,FLNW17,CGMV17] AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  10. MPC for small number of parties Ø Efficiency and Simplicity [MRZ15,AFLNO16,FLNW17,CGMV17] Ø Our focus: MPC with 3 parties AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  11. MPC for small number of parties Ø Efficiency and Simplicity [MRZ15,AFLNO16,FLNW17,CGMV17] Ø Our focus: MPC with 3 parties Ø Corruption : honest majority AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  12. MPC for small number of parties Ø Efficiency and Simplicity [MRZ15,AFLNO16,FLNW17,CGMV17] Ø Our focus: MPC with 3 parties Ø Corruption : honest majority q Majority of the parties are honest q 3PC – at most 1 corruption AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  13. MPC for small number of parties Ø Efficiency and Simplicity [MRZ15,AFLNO16,FLNW17,CGMV17] Ø Our focus: MPC with 3 parties Ø Corruption : honest majority Ø Outsourced Computation AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  14. MPC for small number of parties Ø Efficiency and Simplicity [MRZ15,AFLNO16,FLNW17,CGMV17] Ø Our focus: MPC with 3 parties Ø Corruption : honest majority Ø Outsourced Computation Ø Pre-processing Model AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  15. MPC for small number of parties Ø Efficiency and Simplicity [MRZ15,AFLNO16,FLNW17,CGMV17] Ø Our focus: MPC with 3 parties Ø Corruption : honest majority Ø Outsourced Computation Ø Pre-processing Model Pre-processing phase § AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  16. MPC for small number of parties Ø Efficiency and Simplicity [MRZ15,AFLNO16,FLNW17,CGMV17] Ø Our focus: MPC with 3 parties Ø Corruption : honest majority Ø Outsourced Computation Ø Pre-processing Model Pre-processing phase § q Data-independent Computation q Relatively slow and expensive AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  17. MPC for small number of parties Ø Efficiency and Simplicity [MRZ15,AFLNO16,FLNW17,CGMV17] Ø Our focus: MPC with 3 parties Ø Corruption : honest majority Ø Outsourced Computation Ø Pre-processing Model Pre-processing phase § Online Phase § AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  18. MPC for small number of parties Ø Efficiency and Simplicity [MRZ15,AFLNO16,FLNW17,CGMV17] Ø Our focus: MPC with 3 parties Ø Corruption : honest majority Ø Outsourced Computation Ø Pre-processing Model Pre-processing phase § q Minimized communication Online Phase § q Blazing fast AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  19. BLAZE PROTOCOL AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  20. BLAZE Protocol S 0 S 1 S 2 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  21. BLAZE Protocol S 0 S 1 S 2 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  22. BLAZE Protocol S 0 S 1 S 2 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  23. BLAZE Protocol S 0 S 1 S 2 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  24. BLAZE Protocol S 0 S 1 S 2 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  25. BLAZE Protocol S 0 S 1 S 2 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  26. BLAZE Protocol Communication Cost per !"#$: &. ( Multiplication Gate (malicious) BLAZE : https://eprint.iacr.org/2020/042 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  27. BLAZE Protocol Pre-processing Online Ref Security (#elements) (#elements) Araki et al’17 12 9 Abort Communication Cost per !"#$: &. ( Multiplication Gate (malicious) BLAZE : https://eprint.iacr.org/2020/042 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  28. BLAZE Protocol Pre-processing Online Ref Security (#elements) (#elements) Araki et al’17 12 9 Abort ASTRA 21 4 Fair Communication Cost per !"#$: &. ( Multiplication Gate (malicious) BLAZE : https://eprint.iacr.org/2020/042 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  29. BLAZE Protocol Pre-processing Online Ref Security (#elements) (#elements) Araki et al’17 12 9 Abort ASTRA 21 4 Fair Boneh et al’19 0 3 Abort Communication Cost per !"#$: &. ( Multiplication Gate (malicious) BLAZE : https://eprint.iacr.org/2020/042 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  30. BLAZE Protocol Pre-processing Online Ref Security (#elements) (#elements) Araki et al’17 12 9 Abort ASTRA 21 4 Fair Boneh et al’19 0 3 Abort BLAZE 3 3 Fair Communication Cost per !"#$: &. ( Multiplication Gate (malicious) BLAZE : https://eprint.iacr.org/2020/042 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  31. Privacy Preserving Machine Learning (PPML) Model Query Parameters Result ML Algorithm Alice Bob (Model Owner) (Client) Privacy ?? AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  32. Privacy Preserving Machine Learning (PPML) Model Query Parameters Result ML Algorithm Alice Bob (Model Owner) (Client) Query AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  33. Privacy Preserving Machine Learning (PPML) Model Query Parameters Result ML Algorithm Alice Bob (Model Owner) (Client) Model Parameters AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  34. Solution ?? MPC ML AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  35. MPC MEETS ML AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  36. Privacy Preserving Machine Learning (PPML) Model Query MPC Parameters Result ML Algorithm Alice Bob (Model Owner) (Client) Use MPC to achieve privacy AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  37. MLaaS (3PC Servers) Query Model Parameters Result Alice Bob (Model Owner) (Client) SECURE OUTSOURCED SETTING (SOC) 26-02-2020 AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC

  38. Logistic Regression Neural Networks Linear Regression ML ALGORITHMS CONSIDERED AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

  39. PPML using MPC: Hurdles to Clear Secure Dot Product AJITH SURESH | CRYPTOGRAPHY AND INFORMATION SECURITY LAB ,CSA, IISC 26-02-2020

Recommend


More recommend