Neural Network Name Generation via Adversarial Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating appropriate names for objects such as businesses, products, and research papers can be time-consuming and often requires expertise, leading to either additional effort or outsourcing, which increases costs.
Innovation Solution
Training neural networks for name generation using a discriminator and generator network, where the discriminator network is trained with a target set of words and the generator network produces names that meet specific criteria, guided by metadata for style and properties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual name generation is used, then name quality and relevance are improved, but time consumption and cost increase
Solution Approach 1:
The system enables self-service name generation through a neural network model that automatically creates names without requiring manual intervention from experts. The generator network takes object type and guiding metadata as input and produces names autonomously, eliminating the need for human experts to manually create names while maintaining quality through the trained model's knowledge of naming patterns and conventions
Solution Approach 2:
The patent replaces the mechanical process of manual name generation with an automated neural network system. The generator network substitutes human creativity and expertise with a computational model that uses deep learning to generate names, thereby reducing time consumption while maintaining name quality through the model's training on extensive naming data
2Reliability
If manual name generation is used, then name relevance to object type is improved, but expertise requirements and cost increase
Solution Approach 1:
The system eliminates the need for expert involvement by implementing self-service name generation. The neural network model autonomously understands object types and generates relevant names without requiring users to provide detailed specifications or for experts to review and approve names, thereby reducing expertise requirements while maintaining relevance through the model's inherent understanding of semantic relationships
Solution Approach 2:
The patent changes the parameters of name generation from manual expert judgment to automated model-based generation. The generator network uses guiding metadata parameters (such as style, tone, and characteristics) to control name generation, replacing the need for expert knowledge with programmable parameters that can be adjusted without human intervention
3Productivity
If automated name generation is implemented, then productivity is improved, but system complexity increases
Solution Approach 1:
The system segments the name generation task into distinct functional components: a generator network for creating names, a discriminator network for evaluating quality, and a user interface for interaction. This segmentation allows each component to be optimized independently and simplifies the overall system architecture by dividing the complex name generation process into manageable modules that work together cooperatively
Solution Approach 2:
The patent introduces guiding metadata as an intermediary between the user's object type description and the generator network. This intermediary layer translates high-level object type information into specific generation parameters, simplifying the interface between the user and the complex neural network model while maintaining high productivity through automated generation
Data Source
AI summary
Systems, device and techniques are disclosed for training neural networks for name generation. A target set including words associated with an object type may be received. A discriminator network and a generator network may be trained. The discriminator network may be trained with a training data set that is based on the target set and the generator network may be trained with random inputs and the discriminator network. The discriminator network may be trained for two epochs for each epoch for which the generator network is trained. The generator network may generate words.