Automated Neural Network Design via Data-Driven Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Designing a neural network for learning tasks is challenging for users with limited knowledge, as it requires selecting the appropriate loss function and network structure, which can be daunting without prior expertise.
Innovation Solution
A program that generates a network for inference and learning based on specified data sets, allowing users to input data for learning and perform network training, thereby simplifying the network design process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a designer manually selects loss functions and network structures based on expertise, then the network design achieves high accuracy and appropriateness, but the design process becomes complex and time-consuming for users with limited knowledge
Solution Approach 1:
The system enables automatic network design where the computer itself performs the selection of loss functions and network structures based on the specified data set, eliminating the need for manual expert intervention. The generating unit automatically creates appropriate network configurations by analyzing the data characteristics and selecting suitable architectures and loss functions.
Solution Approach 2:
The patent introduces an intermediary system (the generating unit) that mediates between the user's simple data set specification and the complex network design requirements. This intermediary automatically translates high-level specifications into detailed network configurations, including loss function selection and architectural decisions.
2Ease of operation
If automatic network generation is implemented, then ease of use improves for users with limited knowledge, but the need for expert knowledge in selecting loss functions and network structures is reduced
Solution Approach 1:
The system enables automatic network design where the computer itself performs the selection of loss functions and network structures based on the specified data set, eliminating the need for manual expert intervention. The generating unit automatically creates appropriate network configurations by analyzing the data characteristics and selecting suitable architectures and loss functions.
3Reliability
If manual network design is performed with expert knowledge, then the design quality is high, but the time required for design increases
Solution Approach 1:
The system pre-establishes a database of network configurations, loss functions, and design patterns that have been previously optimized by experts. When a user specifies a data set, the generating unit quickly retrieves and adapts pre-vetted configurations rather than designing from scratch, significantly reducing design time while maintaining quality.
Solution Approach 2:
The system creates network designs by copying and adapting proven configurations from the database rather than performing manual design iterations. The generating unit replicates successful network architectures and loss function combinations that have been previously validated, enabling rapid deployment of high-quality networks.
Data Source
AI summary
The present technology relates to a program, an information processing method, and an information processing apparatus that make it possible to easily design network.A program according to one aspect of the present technology is a program causing a computer to serve as: a generating unit that generates, in response to specification of a data set including data for learning, a network for inference execution and a network for learning corresponding to the data set; and a learning unit that inputs the data for learning to the network for learning and performs learning of the network for inference execution.The present technology is applicable to a program that supports network design by Deep Learning.