Blockchain Model Register for AI Accuracy and Fairness Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence models lack transparency and methods to accurately determine their accuracy over time, especially when versions change and access is restricted, due to their complexity and the obscurity of weight and determination processes.
Innovation Solution
Utilizing blockchain technology to create a digital register that stores identifiers such as hashes of model versions, inputs, and outputs, ensuring transparency and immutability while overcoming storage limitations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If blockchain technology is used to store artificial intelligence model data, then transparency and immutability are improved, but storage capacity is exceeded due to large data volumes
Solution Approach 1:
The patent segments model data into two categories: metadata (stored on-chain in the digital register) and actual model artifacts (stored off-chain). This segmentation allows the blockchain to maintain transparency and immutability for critical verification data without being overwhelmed by large model file storage requirements.
Solution Approach 2:
The patent introduces an intermediary storage layer (off-chain storage system) that bridges the blockchain and model data. The digital register on the blockchain stores cryptographic hashes and metadata that point to off-chain storage locations, allowing verification without direct storage of large model files on the blockchain.
2Measurement precision
If model versions are updated over time, then model accuracy may improve, but tracking and verifying accuracy changes becomes more difficult
Solution Approach 1:
The patent performs preliminary action by storing cryptographic hashes of model versions, inputs, and outputs in the digital register before accuracy verification is needed. This pre-stored hash data enables straightforward comparison and tracking of accuracy changes across versions without complex real-time analysis.
Solution Approach 2:
The patent implements feedback by systematically recording accuracy metrics and comparing them across different model versions in the digital register. This creates a feedback loop where each version's performance is measured against previous versions, enabling continuous accuracy improvement tracking.
3Reliability
If a party controls the model privately, then model security is maintained, but accuracy verification by other parties becomes impossible
Solution Approach 1:
The patent creates cryptographic copies (hashes) of the model data that can be stored and verified on the blockchain without requiring access to the actual model files. These hash copies serve as verifiable proofs of model integrity and accuracy metrics while maintaining the security of the original private model.
Data Source
AI summary
Methods and systems described herein relate to the creation of a digital repository of artificial intelligence models that allows users to determine their individual fairness metric. More specifically, the methods and systems provide this digital repository by storing it on a blockchain network and tracking any changes made to the model and/or its fairness metric.