Secure Multi-Party Computation for Neural Network Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current secure multi-party computation (MPC) methods face challenges in scalability, computational efficiency, and user-friendliness for deep learning (DL) applications, particularly in neural network inference, due to intensive computations, communication overhead, and the requirement for cryptographic expertise.
Innovation Solution
An efficient and automated system for neural network secure MPC inference is introduced, featuring innovative cryptographic primitives and a user-friendly API that supports all operating systems and DL frameworks like TensorFlow, Keras, and PyTorch, enabling secure prediction tasks without revealing model or data ownership, and supporting additional layers and functions not handled by existing protocols.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If existing prediction services require the model provider to install their proprietary model on the user's device, then the user can run advanced analysis and predictions, but the model's intellectual property is revealed
Solution Approach 1:
The patent introduces a secure multi-party computation framework where a trusted third party or secure computation environment acts as an intermediary. This allows the model provider to host the proprietary model while the user provides data, with the computation performed in a secure environment that prevents both parties from accessing the other's sensitive information, thus resolving the IP leakage issue while maintaining prediction capability
Solution Approach 2:
The patent segments the prediction service into separate functional components: model hosting, data provision, secure computation execution, and result delivery. Each component is handled by different parties with limited access rights, allowing the model to remain on the provider's side while data stays on the user's side, preventing intellectual property revelation
2Ease of operation
If existing prediction services require the user to upload their input data to the provider's service, then the model can be applied to user data, but data privacy is harmed
Solution Approach 1:
The patent employs a secure computation intermediary that enables the model to process user data without actually transferring or storing the sensitive data on the provider's side. The intermediary facilitates computation over encrypted or shared data representations, allowing model application while maintaining data privacy and preventing data leakage to the model provider
Solution Approach 2:
The patent separates data storage and computation functions into different locations and access scopes. User data remains on the user's device or in their secure environment, while only necessary computation results are exchanged through the secure framework, preventing data privacy violation while enabling model application
3Loss of information
If secure multi-party computation is used for neural network inference, then data privacy is maintained, but execution time increases due to intensive computations and communication overhead
Solution Approach 1:
The patent implements preliminary actions by pre-computing and caching secure computation results, model parameters, and communication protocols before actual inference runs. This preparation phase reduces the computational and communication overhead during real-time inference, thereby decreasing execution time while maintaining the data privacy benefits of secure multi-party computation
Solution Approach 2:
The patent optimizes execution time by dynamically adjusting security parameters and computation precision levels based on the specific inference workload. This allows the system to maintain adequate data privacy protection while reducing computational intensity and communication overhead, resulting in faster execution times for neural network inference
4Reliability
If secure multi-party computation protocols are implemented, then data privacy and security are maintained, but the system becomes less user-friendly and requires cryptographic expertise
Solution Approach 1:
The patent implements self-service by providing automated secure computation frameworks that handle cryptographic operations, key management, and protocol execution automatically. The system serves itself by managing security complexities internally, allowing users to simply provide data and models without needing cryptographic expertise, thus maintaining reliability while improving ease of operation
Solution Approach 2:
The patent introduces user-friendly intermediary layers and abstraction interfaces that shield users from cryptographic complexity. These intermediaries handle the secure computation protocols automatically, presenting simple data input and result output interfaces to users, thereby maintaining strong data privacy and security while making the system accessible to users without cryptographic knowledge
Data Source
AI summary
A system and method are disclosed for secure multi-party computations. The system performs operations including establishing an API for coordinating joint operations between a first access point and a second access point related to performing a secure prediction task in which the first access point and the second access point will perform private computation of first data and second data without the parties having access to each other's data. The operations include storing a list of assets representing metadata about the first data and the second data, receiving a selection of the second data for use with the first data, managing an authentication and authorization of communications between the first access point and the second access point and performing the secure prediction task using the second data operating on the first data.