Customizable Data Filtering Using Sequential Binary Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data classification methods are inflexible and require retraining of classifiers to adjust sensitivity to specific classes, limiting user customization and efficiency in classifying data.
Innovation Solution
A device and method utilizing multiple independent machine learning classifiers with predefined rules, allowing for customizable data filtering by applying different classification rules and methods at each stage, enabling adjustment of confidence thresholds without retraining, and using binary classifiers to optimize classification steps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single pre-trained classifier is used with fixed rules and thresholds, then the classification system is simple and fast, but it cannot be adjusted to change sensitivity to specific classes without retraining
Solution Approach 1:
The patent divides a single classification task into multiple sequential binary classifiers, each handling a specific decision point. This segmentation allows independent adjustment of each classifier's threshold without affecting the others, enabling customization of sensitivity to specific classes while maintaining overall system functionality.
Solution Approach 2:
The patent introduces dynamic threshold adjustment capability where each binary classifier can have its threshold independently modified based on user needs. This dynamic parameter adjustment allows the system to adapt sensitivity without retraining, transforming a static classification system into a flexible one.
2Adaptability or versatility
If confidence thresholds are adjusted to change sensitivity, then customization is enabled, but the classifier must be completely retrained which is time-consuming
Solution Approach 1:
The patent implements dynamic threshold parameters that can be adjusted without retraining the classifiers. Each binary classifier maintains its trained weights and biases but allows runtime modification of decision thresholds, enabling instant adaptation to different sensitivity requirements without the time-consuming retraining process.
Solution Approach 2:
The patent changes the approach from modifying model parameters (which requires retraining) to modifying decision parameters (thresholds) that control how trained models are applied. This separation allows independent adjustment of sensitivity without affecting the trained classification boundaries.
3Adaptability or versatility
If multiple binary classifiers are used in sequence, then customization and flexibility are improved, but the device complexity increases
Solution Approach 1:
The patent segments the classification problem into a series of binary decisions, each handled by a simple two-class classifier. While this increases the number of components, each component remains computationally simple and can be independently configured, making the complexity manageable and the system highly adaptable.
Data Source
AI summary
There is provided a device and method for classifying data. The device comprises a controller configured to receive data, classify the data into a first class or a second class using a first machine learning classifier, and if the data is classified into the second class, classify the data into one of a third class and a fourth class using a second machine learning classifier. The first and second machine learning classifiers have their own predefined sets of rules for classifying data.