Segmented Model Transmission for Terminal Resource Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing communication networks face challenges in efficiently distributing deep learning models due to high data volume and computing power requirements, leading to increased transmission load and resource shortages, particularly in terminals with insufficient capabilities.
Innovation Solution
The method involves segmenting models into blocks based on terminal needs and computing power, optimizing model distribution by selecting and transmitting only necessary segments, reducing communication overheads and computational demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the entire deep learning model is transmitted to terminals, then model precision and completeness are improved, but transmission load and communication overheads increase significantly
Solution Approach 1:
The patent segments the deep learning model into multiple model segmentation blocks based on terminal computing power and communication resources. Each terminal receives only the specific blocks it needs, rather than the entire model, thereby reducing transmission load while maintaining necessary model functionality.
Solution Approach 2:
Different terminals receive different portions of the model based on their local capabilities. The network device determines which model blocks to transmit to each terminal according to terminal-specific computing power and communication resource conditions, optimizing the balance between model precision and transmission load for each local context.
2Adaptability or versatility
If more model blocks are transmitted to terminals, then model functionality is improved, but computing power requirements and resource consumption increase
Solution Approach 1:
The patent dynamically adjusts the model configuration for each terminal based on real-time assessment of computing power and communication resources. The network device determines the appropriate number and type of model blocks to transmit to each terminal, creating a dynamic adaptation rather than a static one-size-fits-all approach.
Solution Approach 2:
The system changes key parameters including the number of model blocks transmitted, the size of each block, and the transmission schedule, all optimized according to terminal capabilities. This parameter optimization ensures adequate model functionality while controlling computing power consumption and resource usage.
3Productivity
If model segmentation is implemented, then transmission efficiency is improved, but system complexity increases
Solution Approach 1:
The network device performs preliminary assessment of terminal computing power and communication resources before model transmission. It pre-determines the optimal segmentation and allocation of model blocks to each terminal, avoiding complex real-time adjustments during transmission and simplifying the overall system operation.
Data Source
AI summary
A model transmission method includes: in response to receiving at least one model distribution request, determining a first model and obtaining a first quantity of model segmentation blocks by segmenting the first model, where each model distribution request in the at least one model distribution request corresponds to the model segmentation block.