Client-Side Machine Learning with Encrypted Inputs to Reduce Server Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Server systems face significant computational burdens due to the execution of large numbers of machine learning models for transaction risk evaluation, leading to resource utilization and user delay issues.
Innovation Solution
Execute machine learning models on client devices using encrypted input data, leveraging homomorphic encryption to protect model integrity and reduce server load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are executed on the server system, then security assessment and malicious activity identification are improved, but server computational burden and resource utilization deteriorate
Solution Approach 1:
The patent extracts the machine learning model execution from the server system and relocates it to client devices. The server system sends the machine learning model to the client device, where it is executed locally using the device's computational resources. This extraction resolves the contradiction by maintaining security assessment functionality while eliminating the server's computational burden of running these models.
Solution Approach 2:
The patent segments the system into two functional parts: the server system that manages and distributes machine learning models, and client devices that execute these models locally. This segmentation allows the security assessment function to be distributed across multiple client devices rather than concentrated on the server, thereby improving security assessment capability while reducing server resource utilization.
2Reliability
If machine learning models are executed on the server system, then malicious activity identification is improved, but user delay increases
Solution Approach 1:
The patent extracts the time-consuming machine learning model execution from the server system and moves it to client devices. By executing models locally on the client device, the processing time is eliminated from the server-response cycle, thereby maintaining malicious activity identification accuracy while significantly reducing user-perceived delay.
Solution Approach 2:
The patent implements preliminary action by pre-loading and caching machine learning models on client devices before they are needed for security assessments. The server system can send models to client devices in advance, and client devices can cache multiple models locally. When security assessment is needed, the pre-cached models are immediately available for execution, eliminating wait time and reducing user delay.
3Use of energy by stationary object
If machine learning models are sent to client devices, then server computational load is reduced, but model security and integrity protection requirements increase
Solution Approach 1:
The patent applies parameter changes by transforming the machine learning model into a different format or representation that is suitable for secure distribution and execution on client devices. This may involve converting models to serialized formats, using model compression techniques, or applying obfuscation methods. These parameter changes enable the model to be safely transmitted and executed on client devices while maintaining security and integrity, thereby reducing server computational load without proportionally increasing protection complexity.
Data Source
AI summary
Techniques are disclosed relating to the execution of machine learning models on client devices, particularly in the context of transaction risk evaluation. This reduces computational burden on server systems. In various embodiments, a server system may receive, from a client device, a request to perform a first operation and select a first machine learning model, from a set of machine learning models, to send to the client device. In some embodiments the first machine learning model is executable, by the client device, to generate model output data for the first operation based on one or more encrypted input data values that are encrypted with a cryptographic key inaccessible to the client device. The server system may send the first machine learning model to the client device and then receive, from the client device, a response message that indicates whether the first operation is authorized based on the model output data.