Blockchain Data Platform for Confidential Machine Learning Model Valuation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a lack of mechanisms to protect the confidentiality of data processed by third parties while correctly assigning value to the data used in building, training, or running machine learning models.
Innovation Solution
A computer-implemented method and system that involves receiving multiple datasets, running multiple instances of a machine learning algorithm to create corresponding models, determining a candidate model based on outcome comparisons with ground-truth output, assigning value to dataset subsets, creating a smart contract with this value and dataset information, and recording it in a blockchain.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data is shared between parties for machine learning model training, then model accuracy and data utility are improved, but data confidentiality and security are compromised
Solution Approach 1:
The patent introduces a blockchain-based intermediary system that mediates between data providers and machine learning model trainers. The blockchain records data usage, attribution, and value distribution without exposing the actual data contents, thus enabling model training while preserving data confidentiality through cryptographic mechanisms and decentralized verification.
Solution Approach 2:
The patent creates cryptographic copies and hashes of data metadata and usage records stored on the blockchain, rather than sharing the actual sensitive data. These copies enable verification of data usage and attribution without revealing the underlying confidential information, resolving the contradiction between data utility and security.
2Measurement precision
If multiple data sets are contributed to machine learning models, then model performance is improved, but determining data value and attribution becomes complex
Solution Approach 1:
The patent implements a feedback mechanism where the blockchain system continuously tracks which data sets contribute to model improvements, measures their individual impact on model performance, and automatically adjusts value attribution and compensation accordingly. This automated feedback loop simplifies the complex task of multi-dataset value determination.
Solution Approach 2:
The patent creates a universal blockchain-based attribution system that handles multiple data sets, different data types, and various contribution scenarios through a single standardized framework. This multi-functional system simplifies complexity by providing unified data tracking, valuation, and distribution mechanisms applicable to diverse machine learning scenarios.
3Productivity
If data is processed by third parties, then machine learning model training is enabled, but data security and confidentiality protection mechanisms are lacking
Solution Approach 1:
The patent establishes preliminary blockchain-based security protocols, smart contracts, and cryptographic protections before data is shared with third parties for machine learning training. These pre-configured mechanisms ensure data security, track usage rights, and define attribution rules in advance, enabling trusted third-party processing while maintaining confidentiality.
Data Source
AI summary
A plurality of datasets may be received. Multiple instances of a machine learning algorithm may be run to create corresponding multiple machine learning models trained for a specific task in a given domain. Each of the multiple instances may use a different subset of the plurality of datasets in training the corresponding machine learning model. The multiple machine learning models may be run with input data. The multiple machine learning models produce corresponding multiple outcomes. A candidate machine learning model may be determined based on comparing each of the multiple outcomes with ground-truth output. A value associated with the different subset of the plurality of datasets may be determined based on comparing of each of the multiple outcomes with ground-truth output. A smart contract may be created which may include the value and the different subset of the plurality of datasets. The smart contract may be recorded in a blockchain.