AI Model Accuracy Thresholding for Lower-Cost Multi-Model Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI systems combining multiple deep learning models face challenges in accurately determining the minimum accuracy required for each model, leading to excessive computational resources and costs due to over-enhanced model quality.
Innovation Solution
A method for quantitatively analyzing the needed accuracy of deep learning models within an AI system by creating accuracy reference information in a multidimensional space using discrete threshold value evaluations, allowing for efficient training and cost reduction by optimizing computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the quality of deep learning models is over-enhanced to ensure sufficient AI system accuracy, then the recognition accuracy and reliability are improved, but the computational costs and resources increase excessively
Solution Approach 1:
The patent changes the parameter of model accuracy from a fixed high standard to variable thresholds. By establishing multiple accuracy thresholds (first threshold for individual models, second threshold for system output) and allowing models to operate at different accuracy levels based on their contribution to the final result, the system reduces unnecessary computational expenditure while maintaining reliable overall performance.
Solution Approach 2:
Instead of requiring all deep learning models to exceed a high accuracy threshold, the patent applies partial action by setting lower thresholds for individual models whose contributions are less critical to the final output. This allows the system to achieve sufficient overall accuracy without the excessive computational cost of optimizing every component to maximum performance.
2Manufacturing precision
If multiple deep learning models are trained separately to ensure individual model quality, then the manufacturing precision of each model is improved, but the device complexity and training time increase
Solution Approach 1:
The patent segments the accuracy evaluation process into distinct levels: individual model evaluation against a first accuracy threshold and system-level evaluation against a second accuracy threshold. This segmentation allows each model to be trained and evaluated independently with appropriate thresholds, reducing the complexity of coordinating simultaneous optimization across all models while ensuring sufficient precision where needed.
3Measurement precision
If high accuracy thresholds are set for all deep learning models, then the measurement precision of model performance is improved, but the productivity and processing time decrease
Solution Approach 1:
The patent introduces dynamic threshold selection based on model characteristics and system requirements. Different accuracy thresholds are applied dynamically to different models depending on their role in the system, rather than using a static high threshold for all models. This dynamic approach maintains precise measurement of model performance where necessary while accelerating training and evaluation for models where lower thresholds are sufficient.
Data Source
AI summary
A non-transitory computer-readable recording medium storing an analysis program that causes a computer to execute a process, the process includes combining an artificial intelligence (AI) system with a plurality of deep learning models; and creating accuracy reference information for the plurality of deep learning models in a space in which accuracy evaluation information is projected in multiple dimensions, by using discrete threshold value evaluations, for the AI system.