NECSTSpecialTalk - Private Machine Learning with Faster Integer Arithmetic over TFHE
Alex Veidenbaum
Professor of Computer Science at University of California, Irvine
DEIB - NECSTLab Meeting Room (Bld. 20)
On Line via Facebook
July 17th, 2023
10.30 am
Contacts:
Cristina Silvano
Research Line:
System architectures
Professor of Computer Science at University of California, Irvine
DEIB - NECSTLab Meeting Room (Bld. 20)
On Line via Facebook
July 17th, 2023
10.30 am
Contacts:
Cristina Silvano
Research Line:
System architectures
Sommario
On July 17th, 2023 at 10.30 am a new appointment of NECSTSpecialTalk on "Private Machine Learning with Faster Integer Arithmetic over TFHE" will take place in DEIB NECSTLab Meeting Room (Building 20).
This talk will be held by Alex Veidenbaum, Professor of Computer Science at University of California, Irvine, invited by Cristina Silvano, Deib Full Professor and Chair of Computer Engineering at DEIB - Politecnico di Milano.
Machine Learning as a Service makes the privacy of both user data and network weights a critical concern. Using Fully Homomorphic Encryption offers a solution for privacy-preserving computation in a cloud environment by allowing computation directly over encrypted data. Torus FHE (TFHE) offers certain advantages over other FHE schemes. However, software TFHE implementations lack efficient ciphertext-ciphertext integer multiplication, which is needed when both input data and weights are encrypted. This work proposes a new way to improve the performance of such multiplication for TFHE by utilizing a version of carry-save addition. This allows parallelism to be utilized, achieving speedups proportional to the bit width of the plaintext integer operands. It also speeds up multi-operand summation, as when a dot-product is computed. The performance improvements observed are 14.8x for the multiply and 45x for a 32-element dot product, both with 16b operands.
The NECSTLab is a DEIB laboratory, with different research lines on advanced topics in computing systems: from architectural characteristics, to hardware-software codesign methodologies, to security and dependability issues of complex system architectures.
This talk will be held by Alex Veidenbaum, Professor of Computer Science at University of California, Irvine, invited by Cristina Silvano, Deib Full Professor and Chair of Computer Engineering at DEIB - Politecnico di Milano.
Machine Learning as a Service makes the privacy of both user data and network weights a critical concern. Using Fully Homomorphic Encryption offers a solution for privacy-preserving computation in a cloud environment by allowing computation directly over encrypted data. Torus FHE (TFHE) offers certain advantages over other FHE schemes. However, software TFHE implementations lack efficient ciphertext-ciphertext integer multiplication, which is needed when both input data and weights are encrypted. This work proposes a new way to improve the performance of such multiplication for TFHE by utilizing a version of carry-save addition. This allows parallelism to be utilized, achieving speedups proportional to the bit width of the plaintext integer operands. It also speeds up multi-operand summation, as when a dot-product is computed. The performance improvements observed are 14.8x for the multiply and 45x for a 32-element dot product, both with 16b operands.
The NECSTLab is a DEIB laboratory, with different research lines on advanced topics in computing systems: from architectural characteristics, to hardware-software codesign methodologies, to security and dependability issues of complex system architectures.
Every week, the “NECSTSpecialTalk” invites researchers, professionals or entrepreneurs to share their work experiences and projects they are implementing in the “Computing Systems”.
Event will hold on line by Facebook.
