On-Device AI Request Processing With Authenticated PEFT Models

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing parameter efficient fine-tuning (PEFT) techniques for large models consume excessive computational resources and time, and there is a need to secure user-specific models from unauthorized use and protect personal information.

Innovation Solution

An electronic device determines the necessity of executing a PEFT model, performs user authentication, and upon success, generates and decrypts a Query Key Value/Weight (QKV/W) matrix to execute the PEFT model, while failing authentication results in a dummy model operation or foundation model execution.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of energy

If PEFT techniques are used to train only some parameters, then computational resources and time are reduced, but security and protection of personal information are compromised

Engineering Contradiction:

Improvecomputational resourcesVSAvoidsecurity

Core Design Contradiction:

Loss of energyVSReliability

Solution Approach 1:

The patent applies preliminary action by performing user authentication before allowing access to or execution of the PEFT model. The system checks whether a user is authenticated prior to enabling PEFT model operations, ensuring that only authorized users can access the fine-tuned model parameters and personal information stored in cache memory.

Inventive Principle:

Principle #10Preliminary action

2Adaptability or versatility

If PEFT model is executed for every user request, then personalization is improved, but unauthorized access and computational overhead increase

Engineering Contradiction:

ImprovepersonalizationVSAvoidunauthorized access

Core Design Contradiction:

Adaptability or versatilityVSObject-affected harmful factors

Solution Approach 1:

The patent introduces an intermediary authentication mechanism between the user request and the PEFT model execution. The authentication module acts as a mediator that verifies user identity and authorization before allowing the PEFT model to process requests, thereby preventing unauthorized access while maintaining personalized service for authenticated users.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Reliability

If authentication is performed before PEFT model execution, then security is improved, but processing time increases

Engineering Contradiction:

ImprovesecurityVSAvoidprocessing time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs authentication as a preliminary action before PEFT model execution, but optimizes the timing and efficiency of this check. By authenticating users in advance and caching authentication results, the system minimizes the time impact while ensuring security is established before any computational resources are allocated to PEFT model processing.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250337572A1Method of processing user request by using on-device ai model and electronic device for performing the same

Publication Date: 2025.10.30 SAMSUNG ELECTRONICS CO LTD

AI summary

A method of processing a user request by an electronic device using a parameter efficient fine-tuning (PEFT) model, the method including: obtaining, by the electronic device, a prompt from a user; determining, by the electronic device, whether execution of the PEFT model is required to process the prompt; based on determining that execution of the PEFT model is not required, executing a foundation model by the electronic device; based on determining that execution of the PEFT model is required, performing user authentication by the electronic device; based on the user authentication being successful, obtaining, by the electronic device, at least one matrix corresponding to the PEFT model; and executing, by the electronic device, the PEFT model using the at least one matrix.