Multi-Task Learning Model Construction via Staggered Search Layers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for constructing multi-task learning models are inefficient and require manual verification, leading to wastage of resources and suboptimal network structure selection.
Innovation Solution
A method and apparatus that automatically construct a multi-task learning model by creating a search space with staggered subnetwork and search layers, sampling candidate paths, and training parameters using sample data to generate a multi-task learning model for hierarchical prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual verification methods are used to select network structures, then model selection accuracy can be maintained, but construction efficiency deteriorates significantly
Solution Approach 1:
The system performs automatic network structure search and model selection through self-service mechanisms. The neural architecture search algorithm autonomously explores the search space, evaluates candidate structures, and selects optimal models without requiring manual verification, thereby maintaining accuracy while dramatically improving construction efficiency
Solution Approach 2:
The patent replaces the manual mechanical verification process with an automated computational system. The neural architecture search algorithm substitutes human experts' manual evaluation with automated computational evaluation, using loss function calculations and gradient-based optimization to objectively assess and select network structures
2Reliability
If manual verification processes are employed, then resource utilization deteriorates due to human involvement, but model selection reliability can be maintained
Solution Approach 1:
The system achieves self-service by automatically performing network structure evaluation and selection. The algorithm autonomously computes loss functions, performs gradient calculations, and selects optimal architectures without human intervention, eliminating manpower consumption while maintaining reliable model selection through objective computational metrics
Solution Approach 2:
The patent substitutes manual verification processes with automated computational mechanisms. The neural architecture search system replaces human experts' subjective evaluation with objective computational metrics including loss function values and gradient-based performance measures, thereby maintaining reliability while eliminating resource waste associated with manual processes
3Device complexity
If simple network structures are used, then construction complexity is reduced, but learning capability deteriorates
Solution Approach 1:
The system dynamically adjusts network structure complexity through the neural architecture search process. Instead of using fixed simple or complex structures, the algorithm dynamically explores the search space and adapts the network architecture to the specific task requirements, achieving optimal balance between complexity and learning capability for each problem
Solution Approach 2:
The patent changes network structure parameters automatically through the search algorithm. The system varies architectural parameters such as layer depths, channel dimensions, and connection patterns during the search process, transforming static network configurations into dynamically optimized structures that achieve high learning capability without excessive complexity
4Productivity
If automated methods are used to construct models, then construction efficiency is improved, but method complexity increases
Solution Approach 1:
The patent segments the model construction process into distinct modular components: search space definition, candidate generation, evaluation metrics, and selection mechanisms. This segmentation allows the complex automated construction process to be broken down into manageable modules, improving efficiency while making the overall system more comprehensible and maintainable
Solution Approach 2:
The neural architecture search system implements universal algorithms that can handle multiple tasks and network types. The same search framework and evaluation mechanisms work across different problem domains, reducing the need for task-specific customization and thereby improving construction efficiency without proportionally increasing method complexity
Data Source
AI summary
This application relates to a method for constructing a multi-task learning model, an electronic device, and a computer-readable storage medium. The method includes: constructing a search space formed between an input node and a plurality of task nodes by arranging a plurality of subnetwork layers and a plurality of search layers in a staggered manner. A search layer in the plurality of search layers is arranged between two subnetwork layers of the plurality of subnetwork layers. The method includes sampling a path from the input node to each task node of the plurality of task nodes through the search space to obtain a candidate path as a candidate network structure; and training a parameter of the candidate network structure according to sample data, to generate the multi-task learning model for performing a multi-task prediction.