AI Transaction Screening with Adversarial Latent Feature Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI models are vulnerable to adversarial attacks where adversaries manipulate input parameters to exploit vulnerabilities, leading to undetected fraudulent transactions or undesired outcomes by maintaining a low risk score.
Innovation Solution
Implement a system with a machine learning decision model and an adversary detection model that identifies adversarial latent features through a self-calibrating outlier detection model, using quantile estimation and moving averages to detect abnormal activations and block transactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a machine learning decision model is used to evaluate transactions, then decision automation and processing speed are improved, but the system becomes vulnerable to adversarial attacks where fraudsters can manipulate inputs to evade detection
Solution Approach 1:
The patent introduces an intermediary component called an 'adversary detection model' that sits between the transaction input and the machine learning decision model. This intermediary analyzes transactions for adversarial characteristics before they reach the decision model, allowing the system to maintain high processing speed while adding a layer of security detection that identifies manipulated inputs attempting to evade fraud detection.
2Adaptability or versatility
If the machine learning model is trained on historical data, then it learns patterns for decision making, but it may overlook rare or novel patterns that adversarial actors exploit
Solution Approach 1:
The system performs preliminary analysis using the adversary detection model before transactions are evaluated by the main decision model. This preliminary action identifies and flags transactions with adversarial characteristics, allowing the system to handle rare or novel patterns that differ from historical training data by detecting anomalies in how inputs are constructed rather than relying solely on learned patterns from past experiences.
3Reliability
If fraudsters probe various combinations of input parameters, then they can find vulnerabilities in the model, but implementing comprehensive monitoring increases system complexity
Solution Approach 1:
The patent extracts the adversarial detection function as a separate, specialized model distinct from the main transaction decision model. This extracted component focuses specifically on identifying manipulated inputs and adversarial patterns, allowing the main decision model to remain relatively simple while security monitoring is handled by the dedicated adversary detection model that analyzes input characteristics for signs of probing or manipulation.
Data Source
AI summary
A computer-implemented system to detect vulnerabilities in artificial intelligence (AI) models, the system comprising a first AI model for calculating a first score for a first transaction based on one or more features extracted from the first transaction and transaction history associated with the first transaction, the first transaction being tagged as potentially adversarial, in response to determining that the first score is in an improbable range based on comparing first attributes associated with the first transaction with second attributes associated with at least a second transaction, the comparison indicating the first transaction has a low likelihood of occurrence; and a second AI model for identifying adversarial transactions, in response to determining that number of plurality of example transactions scored by the first model is sufficient to train the second AI model.