Computing tool risk discovery
AssigneeWells Fargo Bank, N.A.
InventorRobert Lee Posert
Life of this patent
Estimating the term…
Abstract
Risk associated with end-user computing tools can be discovered automatically. Machine learning and other approaches can be employed to automate discovery of risk associated with end-user computing tools. In one instance, a machine learning model can be constructed and fine-tuned through training that can classify end-user computing tools in terms of risk. The risk can be of a particular type, such as financial or reputational risk, and extent, such as high or low. In another instance, end-user computing tools can be subject to automatic clustering and subsequent risk assessment. Mitigation action can be performed to reduce risk associated with high-risk end-user computing tools.
Claims20
A method comprising:
- receiving training data comprising end-user computing tools with risk labels, the end-user computing tools corresponding to software-based tools that perform a mathematical calculation or logical operation;
- determining, using a machine learning model comprising multiple neural networks, where each neural network of the of the multiple neural networks is trained to identify a unique type of risk, a risk type associated with an end-user computing tool;
- adjusting parameter values of the machine learning model based on training data associated with multiple end-user computing tools;
- determining performance of the machine learning model;
- and
- if performance is determined to be below a threshold, tuning parameter values of the machine learning model via back propagation.
A system comprising:
- a processor configured to:
- receive training data comprising end-user computing tools with risk labels, the end-user computing tools corresponding to software-based tools that perform a mathematical calculation or logical operation;
- determine, using a machine learning model comprising multiple neural networks, where each neural network of the of the multiple neural networks is trained to identify a unique type of risk, a risk type associated with an end-user computing tool;
- adjust parameter values of the machine learning model based on training data associated with multiple end-user computing tools;
- determine performance of the machine learning model;
- and
- if performance is determined to be below a threshold, tune parameter values of the machine learning model via back propagation.
A non-transitory computer readable medium comprising program code that, when executed by one or more processors, is configured to cause the one or more processors to perform steps comprising:
- receiving training data comprising end-user computing tools with risk labels, the end-user computing tools corresponding to software-based tools that perform a mathematical calculation or logical operation;
- determining, using a machine learning model comprising multiple neural networks, where each neural network of the of the multiple neural networks is trained to identify a unique type of risk, a risk type associated with an end-user computing tool;
- adjusting parameter values of the machine learning model based on training data associated with multiple end-user computing tools;
- determining performance of the machine learning model;
- and
- if performance is determined to be below a threshold, tuning parameter values of the machine learning model via back propagation.
Dependencies are inferred from the claim text.