Secure Data Routing via Proxy Re-encryption for AI Services
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for secure data routing in communication networks, such as those used for artificial intelligence services, face challenges in maintaining data privacy and security, particularly when intermediary network entities can read sensitive data, leading to potential information leakage.
Innovation Solution
A system comprising a platform controller and a coordinator that manages AI services by obtaining and providing security credentials for re-encrypting data between devices and AI servers, ensuring secure and anonymous communication through proxy re-encryption techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is transmitted in plaintext for efficient routing, then routing efficiency is improved, but data privacy is compromised due to intermediary entities being able to read the data
Solution Approach 1:
Data is encrypted before transmission using the recipient's public key. This preliminary encryption action ensures that even if intermediaries can read the data during transmission, they cannot understand its content without the corresponding private key, thus resolving the contradiction between efficient plaintext routing and data privacy protection
Solution Approach 2:
Public key cryptography acts as an intermediary mechanism that allows data to be transmitted through the network without requiring the recipient to be directly connected to the sender. The public key serves as a mediator that enables secure transmission through multiple intermediaries while maintaining privacy, as only the intended recipient with the private key can decrypt the data
2Object-affected harmful factors
If cryptographic techniques are applied to AI training data, then data privacy is protected, but processing complexity increases
Solution Approach 1:
The cryptographic operations are extracted from the main AI training process and performed separately during data preparation and transmission phases. This separation allows the AI training itself to focus on model learning while cryptographic protection is handled by dedicated security mechanisms, reducing the complexity burden on the training process
Solution Approach 2:
Each network entity manages its own cryptographic keys and performs encryption/decryption operations independently. This self-service approach distributes the computational burden across multiple entities rather than centralizing it, and allows each entity to optimize its cryptographic operations based on its specific requirements and capabilities
3Object-affected harmful factors
If federated learning is used to keep data local, then data privacy is maintained, but model accuracy decreases due to non-IID local data
Solution Approach 1:
Multiple local models trained on non-IID data are merged through aggregation to create a global model. This merging process combines the strengths of different local models, allowing the global model to learn patterns that individual local models cannot capture alone, thus improving overall accuracy while maintaining the privacy benefits of federated learning
Solution Approach 2:
The aggregation process transforms local model parameters (weights and biases) into a global model through weighted averaging or other aggregation functions. This parameter transformation allows the system to leverage diverse local data distributions without requiring centralization, improving accuracy by combining complementary information from multiple sources
4Object-affected harmful factors
If split learning is implemented to train models locally, then data privacy is protected, but scalability is limited due to sequential learning requirements
Solution Approach 1:
The deep learning model is segmented into multiple components that can be distributed across different network entities. This segmentation allows parallel training of different model components by multiple clients simultaneously, rather than requiring sequential learning, thus improving scalability while maintaining the privacy benefits of local training
Solution Approach 2:
The learning process transitions from a single sequential dimension to multiple parallel dimensions by distributing different model components across multiple clients. This dimensional transformation enables concurrent training operations, significantly improving system scalability and throughput while each client still processes data locally for privacy protection
Data Source
AI summary
There is provided a method and system for supporting secure data routing for artificial intelligence services in a communication network. According to embodiments there is provided a system including a platform controller for managing artificial intelligence (AI) services and a coordinator for managing data transmission of the AI services. The platform controller is configured to obtain, from an AI controller, a first security credential used for re-encryption of uplink data for an AI service, wherein the uplink data includes encrypted data from a device and obtain, from the device, a second security credential for re-encryption of downlink data for the AI service, wherein the second downlink data includes encrypted data from an AI server. The platform controller is further configured to provide, to the coordinator, both the first security credential and the second security credential. The coordinator is configured to receive, from the platform controller, both the first security credential and the second security credential and perform re-encryption of the uplink data for the AI service and send re-encrypted uplink data to the AI server. The coordinator is further configured to perform re-encryption of the downlink data for the AI service and send re-encrypted downlink data to the device.