Shared Learned Model Selection for Device-Specific AI Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing learned models require significant cost and effort to be tailored to specific device environments and conditions, limiting their applicability and accuracy, and general-purpose models lack precision while increasing complexity and cost.
Innovation Solution
A server device that stores pre-learned models for various environments and conditions, selects an optimum model based on device data, and performs additional learning to adapt the model to the device's specific needs, reducing complexity and cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a general-purpose learning model is used to cope with various operation environments and conditions, then the model can be applied to various situations, but the accuracy is low in any environment and conditions as compared with the accuracy of the model specialized in the environment and conditions
Solution Approach 1:
The patent segments the learning model into a base model and multiple specialized models corresponding to different operation environments and conditions. The base model provides general capabilities while specialized models handle specific scenarios, allowing the system to select the appropriate model for each situation rather than using a single general-purpose model.
Solution Approach 2:
The patent performs preliminary learning to create multiple specialized models in advance for different operation environments and conditions. These pre-prepared models are stored and can be quickly selected and applied when needed, avoiding the need to train from scratch for each new situation while maintaining high accuracy.
2Adaptability or versatility
If a general-purpose learning model is used to achieve versatility, then the model can handle various operation environments and conditions, but the complexity of the model increases and the amount of information necessary increases, resulting in an increase in operation cost and memory cost
Solution Approach 1:
Instead of creating one large complex general-purpose model, the patent divides the modeling task into a simple base model and multiple specialized models. Each specialized model is tailored to specific conditions rather than attempting to handle all scenarios, reducing the complexity of individual models while maintaining overall versatility through model selection.
3Measurement precision
If learning is performed from zero (from scratch) to achieve accurate determination and classification, then the accuracy is improved, but it takes considerable cost and time
Solution Approach 1:
The patent performs preliminary learning to create specialized models in advance for different operation environments and conditions. When a new situation arises, the system can directly select and apply the pre-prepared specialized model instead of training from scratch, dramatically reducing the time and computational cost while maintaining high accuracy.
Solution Approach 2:
The patent creates specialized models by copying and adapting from a base model rather than training entirely new models from scratch for each situation. This approach leverages the pre-trained features and structures of the base model, significantly reducing the learning time and computational resources required while achieving accurate determination and classification.
4Measurement precision
If a learned model is obtained through specialized learning for each device, then the accuracy is improved, but it is not possible to use the same learned model in devices different in operation environments and conditions, requiring learning again from scratch
Solution Approach 1:
The patent segments the learning model into a reusable base model and device-specific specialized models. The base model contains general knowledge that can be applied across different devices, while specialized models capture device-specific characteristics. This structure allows the base model to be reused across multiple devices while maintaining high accuracy through the addition of specialized components.
Solution Approach 2:
The patent creates a base model with universal applicability that can serve multiple devices with different operation environments and conditions. This base model is designed to handle general patterns and can be adapted to specific devices through the addition of specialized models, making it possible to reuse the same learned model across different devices rather than training entirely new models for each one.
Data Source
AI summary
A server device configured to communicate, via a communication network, with at least one device including a learner configured to perform processing by using a learned model, includes processor, a transmitter, and a storage configured to store a plurality of shared models pre-learned in accordance with environments and conditions of various devices. The processor is configured to acquire device data including information on an environment and conditions from the at least one device, and select an optimum shared model for the at least one device based on the acquired device data. The transmitter is configured to transmit a selected shared model to the at least one device.