Feature Contribution Clustering for Counterfeit Data Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting counterfeit data are limited to specific data types and machine learning algorithms, failing to provide a versatile approach that effectively identifies and addresses counterfeit data across various contexts.
Innovation Solution
An information processing device and method utilizing explainable AI (XAI) to calculate feature contributions, cluster training data unsupervisedly, and compare cluster variability to identify and correct counterfeit samples, enabling detection and removal of such data without limitations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing detection methods are applied to specific data types and machine learning algorithms, then detection capability is improved for those specific cases, but versatility across different data types and algorithms deteriorates
Solution Approach 1:
The patent creates a universal counterfeit data detection method that works across different data types (images, text, audio, etc.) and machine learning algorithms by using model-agnostic feature extraction and clustering. The system extracts features from any trained model without being specific to particular algorithms or data types, enabling broad applicability while maintaining detection effectiveness.
Solution Approach 2:
The patent changes the detection approach from algorithm-specific parameters to general model behavior parameters such as feature contributions and cluster distributions. By monitoring how features contribute to predictions and how samples cluster in feature space, the system adapts to different models and data types through parameter observations rather than algorithm-specific implementations.
2Measurement precision
If traditional counterfeit data detection methods are used, then detection accuracy for known attack patterns is improved, but ability to detect unknown or novel counterfeit data deteriorates
Solution Approach 1:
The patent performs preliminary clustering of training data before model training to establish baseline clusters representing legitimate data patterns. This preliminary action creates reference clusters that can be compared against test data to detect counterfeit samples, enabling detection of both known and novel attack patterns by comparing against the established baseline rather than relying on predefined attack signatures.
Solution Approach 2:
The system uses feedback from cluster comparison results to identify counterfeit data. By continuously comparing feature contributions and cluster positions of new samples against the baseline clusters, the system provides feedback that enables detection of novel counterfeit patterns that deviate from legitimate data clusters, adapting to new attack types without retraining.
Data Source
AI summary
The present disclosure relates to an information processing device, an information processing method, and a program capable of effectively detecting counterfeit data using a more versatile method.A contribution indicating how much each feature in a training dataset contributes to a predicted label output from a trained model is calculated, the training dataset including both a legitimate sample including only legitimate data and a counterfeit sample at least partially including counterfeit data. Then, clustering is executed to classify each sample of the training dataset into a plurality of clusters using unsupervised learning with the contribution as input, and feature variability between the clusters in the result of the clustering is compared to identify a cluster to which the counterfeit sample included in the training dataset belongs. The present technology can be applied to, for example, a machine learning system that generates a fraud detection model.