Blockchain Gradient Verification Smart Contracts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Centralized databases face issues such as single points of failure, dependency on network connectivity, limited data access, and inefficiencies in high traffic scenarios, particularly when verifying complex computations like gradient descent in machine learning processes.
Innovation Solution
A decentralized blockchain network with smart contracts is used for efficient verification of machine learning model training, where a training participant client generates transaction proposals and endorser nodes execute verify gradient smart contracts to provide endorsements without repeating computationally expensive training procedures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If centralized database is used for storing and managing machine learning training data, then data management and control are simplified, but the system suffers from single point of failure and network connectivity dependency
Solution Approach 1:
The patent segments the centralized database into multiple distributed nodes across a blockchain network. Each node maintains a copy of the training data, eliminating the single point of failure while preserving data management capabilities through distributed consensus mechanisms.
Solution Approach 2:
The patent introduces smart contracts as intermediary components that automate data management operations on the distributed blockchain network. These smart contracts handle data access control, validation, and coordination between nodes, maintaining ease of operation while distributing data storage.
2Quantity of substance
If centralized database is used for data storage, then data redundancy is minimized, but data loss recovery becomes difficult and time-consuming
Solution Approach 1:
The patent implements local quality by allowing each distributed node to maintain its own copy of the training data with local storage capabilities. This enables nodes to independently recover data from their local copies without requiring centralized coordination, reducing data recovery time while maintaining appropriate redundancy across the network.
3Reliability
If traditional blockchain verification is used for machine learning gradient calculations, then data integrity is ensured, but verification speed becomes extremely slow due to recomputing complex training procedures
Solution Approach 1:
The patent extracts the computationally expensive gradient calculation verification from the traditional blockchain consensus process. Instead of requiring full recomputation of training procedures, the system separates verification into lighter-weight operations that check gradient properties without retraining models, dramatically improving verification speed while maintaining integrity.
Solution Approach 2:
The patent changes the verification parameters from full model retraining to gradient property checking. By verifying mathematical properties of gradient calculations (such as convergence properties and bounds) rather than recomputing entire training procedures, the system maintains verification accuracy while achieving orders of magnitude speedup.
4Adaptability or versatility
If multiple devices access centralized database simultaneously, then collaboration is enabled, but bottlenecks occur and data access conflicts arise
Solution Approach 1:
The patent segments the centralized data access point into multiple distributed nodes that can simultaneously handle access requests. Each node independently processes transactions, eliminating bottlenecks while maintaining data consistency through blockchain consensus mechanisms.
Solution Approach 2:
The patent implements preliminary action through smart contracts that pre-validate and prepare data access requests before they reach the database nodes. This preliminary validation reduces conflicts and streamlines concurrent access by establishing access rules and permissions in advance.
Data Source
AI summary
An example operation may include one or more of generating, by a training participant client, a plurality of transaction proposals, each of the plurality of transaction proposals corresponding to a training iteration for machine learning model training related to stochastic gradient descent, the machine learning model training comprising a plurality of training iterations, the transaction proposals comprising a gradient calculation performed by the training participant client, transferring the plurality of transaction proposals to one or more endorser nodes or peers each comprising a verify gradient smart contract, executing, by each of the endorser nodes or peers, the verify gradient smart contract; and providing endorsements corresponding to the plurality of transaction proposals to the training participation client.