18182746. TRAINING NEURAL NETWORKS WITH NON-POLYNOMIAL ELEMENTS FOR HOMOMORPHIC ENCRYPTION COMPUTATIONS USING SUB-NETWORKS AND MULTI-LOSS simplified abstract (International Business Machines Corporation)

From WikiPatents
Jump to navigation Jump to search

TRAINING NEURAL NETWORKS WITH NON-POLYNOMIAL ELEMENTS FOR HOMOMORPHIC ENCRYPTION COMPUTATIONS USING SUB-NETWORKS AND MULTI-LOSS

Organization Name

International Business Machines Corporation

Inventor(s)

Itamar Zimerman of Tel-Aviv (IL)

Nir Drucker of Zichron Yaakov (IL)

TRAINING NEURAL NETWORKS WITH NON-POLYNOMIAL ELEMENTS FOR HOMOMORPHIC ENCRYPTION COMPUTATIONS USING SUB-NETWORKS AND MULTI-LOSS - A simplified explanation of the abstract

This abstract first appeared for US patent application 18182746 titled 'TRAINING NEURAL NETWORKS WITH NON-POLYNOMIAL ELEMENTS FOR HOMOMORPHIC ENCRYPTION COMPUTATIONS USING SUB-NETWORKS AND MULTI-LOSS

The abstract of the patent application describes a system with a processor that can handle non-homomorphic encryption (HE)-friendly analytics models containing non-polynomial elements. The processor is capable of training a substitution model by replacing the non-polynomial element with a sub-network that includes a polynomial replacement element.

  • The system includes a processor that can work with non-homomorphic encryption (HE)-friendly analytics models.
  • The processor is able to train a substitution model by replacing non-polynomial elements with a sub-network containing polynomial replacement elements.
  • This innovation allows for more efficient handling of non-polynomial elements in encryption-friendly analytics models.
  • The system enhances the performance and capabilities of encryption-friendly analytics models by utilizing polynomial replacement elements.
  • By training a substitution model, the system can optimize the processing of non-polynomial elements in encryption-friendly analytics models.

Potential Applications: - Secure data analytics - Encrypted machine learning - Privacy-preserving data processing

Problems Solved: - Efficient handling of non-polynomial elements in encryption-friendly analytics models - Enhanced security and privacy in data processing - Improved performance of encrypted machine learning algorithms

Benefits: - Increased data security - Enhanced privacy protection - Improved efficiency in encrypted data processing

Commercial Applications: Title: Secure Data Analytics System with Enhanced Encryption Capabilities This technology can be utilized in industries such as finance, healthcare, and cybersecurity for secure data analytics and encrypted machine learning applications. It can also be valuable for organizations handling sensitive data that require privacy-preserving data processing.

Questions about the technology: 1. How does the system ensure the security of data during the training of the substitution model? 2. What are the potential limitations of using polynomial replacement elements in encryption-friendly analytics models?


Original Abstract Submitted

An example system includes a processor to receive a non-homomorphic encryption (HE)-friendly analytics model including a non-polynomial element. The processor is to train a substitution model in which the non-polynomial element of the non-homomorphic encryption (HE)-friendly analytics model is replaced with a sub-network including a polynomial replacement element.

Retrieved from "http://wikipatents.org/index.php?title=18182746._TRAINING_NEURAL_NETWORKS_WITH_NON-POLYNOMIAL_ELEMENTS_FOR_HOMOMORPHIC_ENCRYPTION_COMPUTATIONS_USING_SUB-NETWORKS_AND_MULTI-LOSS_simplified_abstract_(International_Business_Machines_Corporation)&oldid=235970"