Learned Model Selection via Generation Environment Metadata
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Selecting a suitable learned model from a plurality of models for user-side needs is challenging due to the influence of the environment where the learning data was acquired and the diverse functions of the models, making it difficult to find a model that meets specific requirements for image recognition and sound recognition tasks.
Innovation Solution
A method and system for providing learned models that involve saving multiple models with associated information such as function, generation environment, and resource requirements, and selecting the most suitable model based on user-side needs information, including use purpose, performance, and resource capabilities, using a server device connected to a user-side device via a network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple learned models are saved in database with diverse functions and generation environments, then the versatility and adaptability of the system is improved, but the complexity of selecting the suitable model increases
Solution Approach 1:
The patent introduces model information as an intermediary element that mediates between the diverse learned models and the user needs. This model information includes function descriptions and generation environment details, serving as a bridge to facilitate accurate matching without increasing system complexity. The intermediary structure allows the system to handle versatile model selection while maintaining manageable complexity through organized information representation.
Solution Approach 2:
The patent applies preliminary action by pre-storing model information alongside each learned model in the database. This includes pre-documented function characteristics and generation environment parameters. When a selection request arrives, the system can immediately compare pre-prepared model information against user needs without performing complex real-time analysis, thus resolving the contradiction between versatility and selection complexity.
2Measurement precision
If the system stores detailed model information including function and generation environment, then the measurement precision of model matching is improved, but the loss of information storage increases
Solution Approach 1:
The patent extracts only the essential matching criteria from complete model information - specifically function descriptions and generation environment parameters. Rather than storing all possible model attributes, the system extracts and stores only those elements necessary for accurate matching against user needs. This extraction approach maintains high measurement precision for model selection while minimizing the quantity of stored information.
3Reliability
If the system performs comprehensive matching between sensor metadata and application metadata, then the reliability of model selection is improved, but the productivity of the system decreases
Solution Approach 1:
The patent applies preliminary action by pre-organizing model information into structured categories (function and generation environment) before the matching process. This pre-organization allows the system to perform reliable comprehensive matching more efficiently, as the information is already arranged for comparison. The preliminary structuring of data reduces the actual processing time during selection, thus maintaining reliability while improving productivity.
Data Source
AI summary
A method for selectively providing a learned model using a learned model providing device includes saving various learned models and model information associated with a generation environment in a database of the learned model providing device. The generation environment indicates an environment where sensing data used for generating each of the learned models is sensed. The method further includes acquiring needs information associated with a use environment of a user side device, and selecting one of the learned model suitable for the needs information by referring to the model information by a processor of the learned model providing device.