Authentication Risk Assessment via Selective Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Adaptive authentication systems face limitations in tuning machine learning models due to the high cost and limited availability of explicit feedback from manual investigations, which restricts the input for risk assessment of authentication requests.
Innovation Solution
A method that receives authentication requests with and without post-authentication feedback, determines their status as genuine or fraudulent, identifies distinctive attributes, assigns these statuses, performs computations to assess risk, and provides computational results for processing authentication requests, using an electronic apparatus with network interface, memory, and control circuitry.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If explicit feedback from manual investigations is used to tune machine learning models, then model accuracy is improved, but investigation cost increases
Solution Approach 1:
The system applies partial action by selectively investigating only a subset of authentication requests rather than all requests. The machine learning model processes the majority of requests automatically, while manual investigations are performed only on selected cases to generate feedback, thereby reducing investigation costs while maintaining model accuracy improvement.
Solution Approach 2:
The system implements feedback by using results from manual investigations to continuously tune and improve the machine learning model. The feedback loop allows the model to learn from investigated cases and improve its predictions, reducing the need for extensive manual investigations over time while maintaining high accuracy.
2Measurement precision
If manual investigations are performed to generate input for machine learning, then model training data quality is improved, but productivity decreases
Solution Approach 1:
The system performs manual investigations on only a partial set of authentication requests rather than all requests. This selective approach ensures that the machine learning model receives high-quality training data from investigated cases while maintaining high authentication processing throughput for the majority of requests that are handled automatically.
Solution Approach 2:
The machine learning model serves itself by automatically processing the majority of authentication requests without manual intervention. The system self-optimizes by using feedback from a small subset of investigated cases to improve its own performance, thereby maintaining high productivity while ensuring data quality.
3Loss of energy
If investigation budget is limited, then resource allocation is optimized, but input data quantity for machine learning is reduced
Solution Approach 1:
The system uses feedback from a strategically selected subset of investigated requests to maximize learning efficiency. By carefully choosing which requests to investigate based on their informational value, the system optimizes the feedback loop to achieve maximum model improvement with minimum investigation budget, effectively compensating for the limited input data quantity.
Solution Approach 2:
The system changes parameters by selectively adjusting which authentication requests are subjected to manual investigation based on risk scores, request characteristics, and other factors. This parameter optimization ensures that the limited investigation budget is allocated to cases that provide the most valuable training data, maximizing input data quality and quantity within budget constraints.
Data Source
AI summary
There is disclosed a technique for use in providing an assessment of authentication requests. In one embodiment, the technique comprises receiving an authentication request with post-authentication feedback and an authentication request with no post-authentication feedback. In the same embodiment, the post-authentication feedback can include a marking indicating that the request is one of a genuine or fraudulent status after review by an analyst. If a request does not possess a post-authentication feedback then it is considered genuine status. The technique can then assign the status of the requests to a distinctive attribute associated with the requests before performing a computation which produces a computational result that is indicative of the risk associated with the distinctive attribute.