Blockchain AI Training via Sub-Model Distribution Across Peers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The communication gap between AI workloads and blockchain platforms leads to inefficiencies, high data transfer costs, data transformation overhead, and security risks, as AI workloads run on one platform while blockchain runs on another, compromising training integrity.
Innovation Solution
A blockchain gateway system divides AI models into sub-models, assigning them to different blockchain peers for training, with real-time monitoring and recording of training results on the blockchain to ensure reliability and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If AI workloads run on a separate centralized platform while blockchain runs on another platform, then platform management is simplified, but communication efficiency deteriorates and data transfer costs increase
Solution Approach 1:
The patent merges AI workloads and blockchain operations into a single integrated platform. The system allows AI models to be trained directly on blockchain nodes, eliminating the need for separate centralized AI platforms. This integration enables direct communication between AI components and blockchain ledgers, improving communication efficiency while maintaining platform manageability through unified architecture.
Solution Approach 2:
The patent introduces an intermediary layer that enables seamless interaction between AI workloads and blockchain operations. This intermediary mechanism allows AI models to access blockchain data and commit training results directly to the ledger without requiring complex cross-platform data transformation, thus improving communication efficiency while keeping platform management straightforward.
2Adaptability or versatility
If data is transferred between separate AI and blockchain platforms, then platform independence is maintained, but data transfer overhead increases and security risks emerge
Solution Approach 1:
The patent combines AI processing and blockchain operations within the same platform environment. AI models are trained directly on blockchain nodes, and training results are committed directly to the blockchain ledger without external data transfer. This eliminates data transfer overhead and associated energy consumption while maintaining system versatility through the integrated architecture.
Solution Approach 2:
The patent creates local copies of blockchain data within the AI processing environment and local copies of AI models within the blockchain node environment. This allows AI workloads to access required data without transferring it across platforms, and blockchain nodes to execute AI training locally, thereby eliminating data transfer overhead while preserving platform independence through virtualized environments.
3Ease of operation
If AI models are trained on a centralized platform, then training management is simplified, but training reliability deteriorates due to security risks
Solution Approach 1:
The patent segments the AI training process into distributed components that run on multiple blockchain nodes. Instead of centralized training management, the system divides the neural network into sub-models distributed across different nodes. Each node independently trains its sub-model and commits results to the blockchain ledger, providing simplified management through automated distributed processes while enhancing reliability through blockchain's immutable record-keeping and consensus mechanisms.
Solution Approach 2:
The patent implements feedback mechanisms where blockchain nodes continuously monitor and verify AI training processes. Training results are committed to the blockchain ledger where they can be verified by consensus mechanisms, providing real-time feedback on training integrity. This feedback loop ensures that any compromised models are detected and prevented from being deployed, maintaining both simplified management and high reliability.
Data Source
AI summary
An example operation may include one or more of dividing a neural network that corresponds to an artificial intelligence (AI) model into a plurality of sub-models, assigning the plurality of sub-models to a plurality of blockchain peers, respectively, training the sub-models, via the plurality of blockchain peers, to generate training results within an iteration, and committing the training results to a blockchain which is accessible by the plurality of blockchain peers.