ML Decision Boundary Validation for Attribute Inference Risk Assessment

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for privacy risk assessment of machine learning models do not adequately consider the structure of the target model, leading to incorrect candidate outputs in attribute inference attacks, resulting in a weaker than expected risk prediction.

Innovation Solution

An information processing apparatus and method that determines the validity of target data based on the decision boundary of a machine learning model, specifically a decision tree, to accurately estimate the value of an unknown attribute by excluding candidate values that match the threshold values of the model's conditional branches, thereby enhancing the accuracy of risk assessment.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If attribute inference attack methods are used without considering the model structure, then the attack can be performed using simple techniques, but the risk assessment accuracy deteriorates due to incorrect candidate outputs

Engineering Contradiction:

Improvesimplicity of attribute inference attackVSAvoidrisk assessment accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The patent changes the parameter of considering model structure by introducing decision boundary information into the attribute inference attack. The determining unit checks whether candidate data satisfies the decision boundary conditions of the machine learning model, thereby modifying the attack method to account for model-specific characteristics and improving risk assessment accuracy.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent replaces the mechanical system of generic attribute inference with a more sophisticated system that incorporates decision boundary validation. The determining unit substitutes simple attribute matching with a check against model-specific decision boundaries, replacing brute-force inference with structure-aware inference.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If candidate data is not filtered based on decision boundary, then the attribute inference process is faster, but incorrect candidates are output reducing prediction accuracy

Engineering Contradiction:

Improvespeed of attribute inferenceVSAvoidcorrectness of inference result

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent applies preliminary action by having the determining unit check whether candidate data satisfies the decision boundary conditions before the estimating unit processes the data. This preliminary validation filters out incorrect candidates early in the process, ensuring only valid candidates are used for attribute estimation and improving result reliability.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The determining unit acts as an intermediary between the candidate data generation and the attribute estimation processes. It mediates by validating candidate data against decision boundary conditions, filtering out incorrect candidates before they reach the estimating unit, and ensuring only reliable candidates proceed to inference.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Reliability

If the determining unit checks decision boundary conditions for all candidate data, then the reliability of inferred attributes improves, but the computational complexity increases

Engineering Contradiction:

Improvevalidity of training dataVSAvoidcomplexity of data validation process

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies segmentation by dividing the data processing into distinct stages: the determining unit first segments and filters candidate data based on decision boundary conditions, and only the valid segmented data is then processed by the estimating unit. This segmentation improves reliability by ensuring valid data is processed while managing complexity through structured processing.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20240354644A1Information processing apparatus, information processing method, and storage medium

Publication Date: 2024.10.24 NEC CORP

AI summary

An information processing apparatus of the present disclosure includes: a determining unit that determines, based on a decision boundary of a machine learning model that relates to a first attribute value and a second attribute value input to the machine learning model and on target data including the pair of a known value of the first attribute value and an unknown candidate value of the second attribute value, whether the target data is valid as training data for the machine learning model; and an estimating unit that estimates the value of the second attribute value from the candidate value of the second attribute value included by the target data determined to be valid.