Classification Model Generation for Microcontroller Memory Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Microcontrollers are limited by their memory and computing resources, making it difficult to implement complex classification models, which often require significant data storage and computational power, thereby restricting their suitability for certain classification tasks.
Innovation Solution
A method for generating a simple and efficient classification model by creating a classification table with defined ranges of values for each characteristic, allowing for reduced memory usage and easy updating, suitable for microcontrollers with limited resources, and avoiding overfitting by assigning classes based on occurrences of learning data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If complex classification models are used to improve classification accuracy, then classification performance is improved, but memory occupancy and computational requirements increase, making them unsuitable for microcontrollers with limited resources
Solution Approach 1:
The patent segments the classification model into a simplified structure with a classification table that divides the feature space into discrete regions. Each region is assigned to a specific class, creating a segmented approach that reduces memory requirements while maintaining classification capability. This segmentation principle directly addresses the contradiction by breaking down complex models into manageable, resource-efficient components suitable for microcontrollers.
Solution Approach 2:
The patent extracts only the essential information needed for classification from the learning data, discarding redundant details. By extracting key features and creating a condensed classification table that maps feature combinations to classes, the model retains classification accuracy while significantly reducing memory occupancy. This extraction principle resolves the contradiction by keeping only the necessary computational elements.
2Reliability
If complex classification algorithms are used to improve classification accuracy, then classification performance is improved, but execution time increases, making them too slow for microcontrollers with limited computing power
Solution Approach 1:
The classification algorithm is segmented into simple lookup operations rather than complex computations. The classification table is pre-computed and stored, allowing the microcontroller to simply query the table during execution. This segmentation transforms complex algorithms into fast, memory-based lookups, resolving the contradiction between accuracy and execution time.
Solution Approach 2:
The patent applies preliminary action by pre-computing and storing the classification table during model generation. The complex computational work is performed in advance when data is available, and the results are stored for quick retrieval during actual classification. This preliminary computation eliminates the need for time-consuming calculations during execution, resolving the contradiction between accuracy and execution time.
3Reliability
If the classification model uses all learning data to improve accuracy, then classification performance is improved, but the model becomes overly fitted to training data, reducing generalization to unseen data
Solution Approach 1:
The patent extracts only the essential classification boundaries and key features from the learning data, discarding redundant details that cause overfitting. By extracting only the necessary information to define meaningful regions and assigning classes based on dominant patterns rather than every individual data point, the model achieves accuracy while maintaining generalization capability. This extraction principle resolves the contradiction by filtering out noise that would cause overfitting.
Data Source
AI summary
A computer-implemented method for generating a classification model includes: obtaining at least one group of learning data, identifying at least one characteristic to be studied of the learning data, extracting a value of each characteristic defined for all learning data, identifying ranges of values for each characteristic from the extracted values, creating a classification table, assigning a class to each cell of the classification table according to a number of occurrences of the learning data of each group according to their extracted value of each characteristic with respect to the ranges of values defined for each studied characteristic, and generating a classification model comprising the classification table.