Extensible Device Architecture for AI Integration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Companies face challenges in developing products with AI technology due to the need for advanced systems and specialized workers, making it difficult for general manufacturers to implement machine learning and incorporate AI capabilities into their devices, especially when sequence control is involved.
Innovation Solution
A device architecture with extensibility, modeled as an ability acquisition model, that includes an ability unit, data input unit, and data output unit, allowing for easy addition of new abilities through ability setting data, input setting data, and output setting data, enabling outsourcing of development and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning is implemented using advanced systems and specialized workers, then AI capabilities can be acquired, but device complexity and manufacturing difficulty increase significantly
Solution Approach 1:
The patent segments the machine learning system into three independent modules: a training system that performs learning offline, a model generation system that creates inference programs, and an embedded inference system in the device. This segmentation allows complex AI training to be separated from simple device deployment, reducing the complexity burden on the end device while maintaining AI capabilities.
Solution Approach 2:
The patent introduces an intermediary model generation system that translates complex training results into simplified inference programs suitable for embedded devices. This intermediary layer acts as a bridge between the advanced training system and the constrained device environment, enabling AI capability transfer without directly implementing the full complexity in the device.
2Measurement precision
If machine learning training systems are built with large-scale computation resources, then learning accuracy improves, but ease of manufacture decreases for general manufacturers
Solution Approach 1:
The patent performs all complex training and model generation actions preliminarily in advance, before the device is manufactured or deployed. The training system pre-processes all computation-intensive tasks and generates ready-to-use inference programs, which are then simply embedded in the device. This preliminary action allows high learning accuracy to be achieved without requiring the manufacturing process to handle complex training operations.
Solution Approach 2:
The patent creates a copy of the trained model in the form of an inference program that can be deployed in the device. Instead of requiring the device to perform complex training operations, the device receives a copied representation of the learned knowledge in the form of pre-processed inference programs, simplifying manufacturing while preserving learning accuracy.
3Reliability
If AI technology is developed by specialized companies, then learning quality improves, but adaptability to specific device applications decreases
Solution Approach 1:
The patent creates a universal model generation system that can translate training results into device-specific inference programs across different application domains. The system is designed to handle multiple types of devices and applications through a common framework, enabling high-quality AI training to be adapted to diverse specific applications without sacrificing either quality or adaptability.
Data Source
AI summary
A device with extensibility includes an architecture modeled as an ability acquisition model including an ability unit for implementing an ability, an data input unit that is an interface for an input from the ability unit, and a data output unit that is an interface for an output from the ability unit, as an architecture for additionally incorporating a new ability to a basic configuration of the device, and includes an ability setting unit for adding the new ability to the device by setting a function to each of the ability unit, the data input unit, and the data output unit, based on ability providing data including ability setting data, input setting data, and output setting data.