Personalized Classifier Trained via Generic Self-Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classifiers face complexity in event categorization due to large variations in sample data among individuals and sensing variations, leading to reduced accuracy in recognizing physical activities and gestures.
Innovation Solution
A personalized classifier is trained using self-training data generated by generic classifiers, tailored to an individual's behavior and movement patterns, implemented as a hardware solution with a programmable pattern recognition neural network, which substitutes generic classifiers for accurate real-time event classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a generic classifier is used for event categorization, then the device can handle multiple individuals and scenarios, but the classification accuracy decreases due to large variations in sample data among individuals
Solution Approach 1:
The classification system is segmented into two distinct components: a generic classifier that handles multiple individuals and a personalized classifier that handles individual-specific variations. This segmentation allows the system to maintain versatility through the generic classifier while achieving high accuracy through the personalized classifier trained on individual sample data.
Solution Approach 2:
The system performs preliminary action by collecting and storing sample data from each individual during an initial setup phase. This pre-collected data is then used to train personalized classifiers before actual event categorization begins, enabling the system to adapt to individual characteristics in advance rather than during real-time operation.
2Measurement precision
If a personalized classifier is trained for each individual, then classification accuracy improves, but the device complexity and training data requirements increase
Solution Approach 1:
The generic classifier serves as an intermediary that generates initial labels for sample data, which are then used to train personalized classifiers. This intermediary approach simplifies the training process by providing pre-labeled data, reducing the complexity of manual annotation while still enabling accurate personalized classification.
Solution Approach 2:
The system creates a personalized copy of the classification model tailored to each individual's characteristics. Instead of modifying the generic classifier, a separate personalized classifier is trained using the individual's sample data, allowing each user to have their own optimized model without affecting others.
3Measurement precision
If more training data is collected for each individual, then the personalized classifier accuracy improves, but the memory requirements and power consumption increase
Solution Approach 1:
The system uses a practical amount of training data that is sufficient to achieve good personalization without requiring excessive data collection. This partial action approach balances the need for accuracy with the constraints of memory and power, avoiding the diminishing returns that would occur with excessive data collection.
Data Source
AI summary
Systems and methods may provide for using one or more generic classifiers to generate self-training data based on a first plurality of events associated with a device, and training a personal classifier based on the self-training data. Additionally, the one or more generic classifiers and the personal classifier may be used to generate validation data based on a second plurality of events associated with the device. In one example, the personal classifier is substituted for the one or more generic classifiers if the validation data indicates that the personal classifier satisfies a confidence condition relative to the one or more generic classifiers.