Inference Network Model Decryption Speed via Data Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for decrypting inference network models in data communication are slow, impacting processing speed, especially when using encrypted models.
Innovation Solution
Separating the inference network model data into two files, one containing layer configuration information and the other coefficient data, and encrypting the first file with stronger encryption, while using weaker encryption for the second file, allowing for faster decryption by starting the process before complete decryption of the first file is completed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the inference network model is encrypted using fully homomorphic encryption, then the security and privacy of data communication is improved, but the decryption processing time increases significantly
Solution Approach 1:
The patent segments the inference network model into two separate files: a first file containing layer configuration information and a second file containing coefficient data. This segmentation allows the system to apply different encryption strategies to different parts of the model, thereby reducing overall decryption time while maintaining security for critical components.
Solution Approach 2:
The patent applies different encryption methods to different parts of the inference network model based on their security requirements and decryption complexity. The layer configuration information in the first file uses encryption that balances security and speed, while the coefficient data in the second file can be processed more quickly. This local differentiation of encryption quality resolves the contradiction between security and decryption speed.
2Measurement precision
If the inference network model is decrypted completely before inference processing, then the accuracy of inference is maintained, but the overall processing speed decreases
Solution Approach 1:
The patent performs preliminary decryption of the first file (layer configuration information) before complete decryption of the entire model is finished. By having the layer configuration ready in advance, the inference apparatus can begin setting up the inference pipeline and performing preliminary processing steps, thereby reducing the total time from data reception to inference completion while maintaining accuracy.
Solution Approach 2:
The patent enables continuous processing by allowing the inference apparatus to start working on inference tasks before the complete model decryption is finished. The layer configuration from the first file can be used immediately to prepare the inference environment, while the coefficient data from the second file is being decrypted in parallel, thus eliminating idle waiting time and maintaining continuous useful action throughout the process.
Data Source
AI summary
An information processing apparatus separates data representing an inference network model into first data including information for representing at least a layer configuration of the inference network model, and second data representing coefficient data of each layer. The apparatus generates a first file including the first data and a second file including the second data, and transmits the first file and the second file to an external apparatus. The apparatus generates the first file from data subjected to encryption processing on the first data.