ML Model Provenance via Artefact Fingerprinting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models are complex and difficult to identify for modifications or duplications due to their large number of artefacts, making it challenging to detect unauthorized or malicious actions.
Innovation Solution
Generating fingerprints for each artefact in machine learning models to create a model indicator, which is then compared to reference models for similarity analysis to determine provenance, allowing for remediation actions such as isolating or disabling the model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are modified or duplicated for malicious purposes, then the model functionality is altered or stolen, but detecting these actions becomes difficult due to the large number of artefacts
Solution Approach 1:
The patent segments the machine learning model into individual artefacts (weights, biases, activations, gradients) and generates unique fingerprints for each artefact. This segmentation enables granular tracking and detection of modifications, transforming the undetectable complex model into monitorable discrete components.
Solution Approach 2:
The patent introduces fingerprints as an intermediary mechanism that bridges the gap between model artefacts and detection systems. These fingerprints serve as mediators that encode artefact characteristics in a comparable format, enabling the detection system to identify modifications without directly analyzing the complex model structure.
2Measurement precision
If fingerprints are generated for each artefact to enable detailed comparison, then detection accuracy is improved, but computational complexity and storage requirements increase
Solution Approach 1:
The patent extracts essential characteristics from each artefact to create compact fingerprints. Instead of storing or processing entire artefacts, only the most discriminative features are captured in fingerprint form, reducing computational complexity while maintaining detection accuracy.
Solution Approach 2:
The patent transforms artefact data into fingerprint representations by changing the parameter format. This transformation converts complex multi-dimensional artefact data into standardized fingerprint parameters that are computationally efficient to store and compare, while preserving the essential information needed for detection.
Data Source
AI summary
Data is received that characterizes artefacts associated with each of a plurality of layers of a first machine learning model. Fingerprints are generated corresponding to each of the artefacts in each layer. The generated fingerprints can collectively form a model indicator for the first machine learning model. A second machine learning model then determines, based on the generated fingerprints, whether the first machine learning model is derived from another machine learning model. Data provided this characterization can be provided to a consuming application or process. This second machine learning model can be trained model with historical fingerprints having a known provenance classification. Related apparatus, systems, techniques and articles are also described.