Neural Network Model Building via Sub-Network Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning-based network model building methods require extensive human intervention and resource consumption, especially when building neural network models from scratch, which leads to low efficiency and high workload due to the need for complex parameter adjustments and labeled data generation.
Innovation Solution
A method and apparatus that analyze the data processing procedure of an original network model, hierarchically build sub-networks using actual operating data as a reference dataset, perform optimization training on these sub-networks, and merge them to form a target network model, allowing for flexible and efficient model building without designing the entire structure from the beginning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a neural network model is built from scratch using traditional methods, then the model can be customized to fit specific needs, but the process requires extensive human intervention, complex parameter adjustments, and labeled data generation, leading to low efficiency and high workload
Solution Approach 1:
The patent divides the original network model into multiple sub-networks, each responsible for specific data processing steps. This segmentation allows the complex model building process to be broken down into manageable units that can be independently optimized and combined, reducing the overall complexity and improving efficiency
Solution Approach 2:
The patent performs preliminary actions by extracting reference datasets from the original network model's operating data before building the new model. This pre-extraction of labeled data and processing procedures eliminates the need for time-consuming manual labeling and allows the sub-networks to be built and trained more efficiently
2Reliability
If extensive labeled data is generated and complex parameter adjustments are performed, then the model performance can be optimized, but the manpower costs and time required for model building increase significantly
Solution Approach 1:
The patent creates sub-networks that copy and adapt the structure and functionality of the original network model's processing steps. By replicating the proven effective components of the original model, the new model inherits their performance characteristics while avoiding the time-consuming process of designing and training them from scratch
Solution Approach 2:
The patent optimizes parameters of the sub-networks by adjusting them based on the reference datasets and specific task requirements. This targeted parameter optimization maintains high model performance while reducing the overall time required compared to training entire models from scratch
3Adaptability or versatility
If the entire network structure is designed from the beginning, then the model can be tailored to specific requirements, but the building process becomes complex and time-consuming
Solution Approach 1:
The patent segments the network model into functional sub-networks that can be independently designed and optimized. This allows the model to be adapted to specific requirements by selecting and combining appropriate sub-networks, making the building process easier compared to designing the entire structure at once
Solution Approach 2:
The sub-networks are designed to perform multiple data processing functions that can be applied to different tasks. This universality allows the same sub-network components to be reused and adapted to various requirements, improving both model adaptability and building ease
Data Source
AI summary
A method and an apparatus to build a machine learning based network model are described. For example, processing circuitry of an information processing apparatus obtains a data processing procedure of a first network model and a reference dataset that is generated by the first network model in the data processing procedure. The data processing procedure includes a first data processing step. Further, the processing circuitry builds a first sub-network in a second network model of a neural network type. The second network model is the machine learning based network model to be built. The first sub-network performs the first data processing step. Then, the processing circuitry performs optimization training on the first sub-network by using the reference dataset.