Distributed Model Inference With Partial Data Sharing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

User data security is compromised during model inference on a server and client due to the need for the server to obtain and process user data, which is not adequately protected.

Innovation Solution

A model inference method where a client splits processing results and sends partial data to a server, using encryption and secure multi-party computation to ensure data security and efficiency, while the client combines results to achieve a complete inference outcome.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the server obtains and processes all user data for model inference, then the inference accuracy is improved, but the data security is compromised

Engineering Contradiction:
Improveinference accuracyVSAvoiddata security
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The user data is divided into multiple parts (first data and second data) that are processed separately by different models (first model on terminal, second model on server). This segmentation allows the server to process only partial data, reducing security risks while maintaining inference accuracy through coordinated processing of all data parts.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

A first processing result is generated as an intermediary output from the first model on the terminal, which then serves as input to the second model on the server. This intermediary structure enables distributed processing where neither party needs access to complete raw data, thus preserving security while achieving accurate inference.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Loss of information

If the server processes all user data, then the inference completeness is improved, but the bandwidth consumption and computation time increase

Engineering Contradiction:
Improveinference completenessVSAvoidcomputation time
Core Design Contradiction:
Loss of informationVSLoss of time

Solution Approach 1:

The computation task is segmented between terminal and server. The first model on the terminal processes data locally to generate a first processing result, reducing the amount of data that needs to be transmitted and processed by the server. This division reduces bandwidth consumption and overall computation time while maintaining inference completeness.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The first model performs preliminary processing of user data on the terminal before data is sent to the server. This preliminary action transforms raw user data into a processed form (first processing result) that requires less computational resources on the server, thereby reducing total computation time and bandwidth usage.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4711986A1Model inference method and device
Publication Date: 2026.03.18 HUAWEI CLOUD COMPUTING TECHNOLOGIES CO LTD
  • EP4711986A1 patent drawingFigure 1
  • EP4711986A1 patent drawingFigure 2
  • EP4711986A1 patent drawingFigure 3

AI summary

A model inference method and apparatus are disclosed, and relates to the field of machine learning technologies. A client and a server use respective deployed models to process different parts of user data, to obtain respective output results. In addition, the client obtains the output result of the server, and obtains an inference result based on the output results of the server and the client. Compared with a case in which the server needs to obtain all the user data in an inference process, in this application, the server obtains only a part of the user data. As the server cannot obtain, based on the part of the user data, all content included in the user data, security of the user data is ensured. In addition, the client needs to send only the part of the user data to the server, so that a bandwidth resource occupied by data transmission between the client and the server and time consumed by the transmission can be reduced, and inference efficiency can be improved.