Classifier Chains With Label Grouping for Multi-Label Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning systems face inefficiencies and inaccuracies in performing multi-label classification due to the computational complexity and label interdependence challenges, particularly when dealing with large label sets and correlated relationships.
Innovation Solution
The use of classifier chains that leverage co-occurrence frequencies and correlations among labels, grouping them into subsets to minimize within-group correlations, optimizing the sequence of classifiers to improve accuracy and reduce computational operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing multi-label classification methods are used, then the system can handle multiple classification labels, but computational inefficiencies and accuracy issues occur when handling large volumes of classification labels
Solution Approach 1:
The patent divides the multi-label classification problem into multiple binary classification sub-problems. Each binary classifier handles a specific label, and the classifiers are arranged in a chain where predictions from previous classifiers inform subsequent ones. This segmentation allows the system to efficiently handle large numbers of labels while maintaining accuracy by breaking down the complex multi-label problem into manageable binary decisions.
2Adaptability or versatility
If a large number of classification labels are processed, then comprehensive classification coverage is achieved, but computational operations and training data needs increase
Solution Approach 1:
The patent segments the classification task into independent binary classifiers, each handling one label. This allows the system to scale to large numbers of labels without proportionally increasing computational complexity, as each binary classifier is computationally simpler than a full multi-label classifier. The modular structure enables efficient processing while maintaining comprehensive label coverage.
Solution Approach 2:
The patent performs preliminary actions by using predictions from earlier classifiers in the chain to inform subsequent classification decisions. This preliminary classification information is reused across multiple binary classifiers, reducing redundant computations and training data requirements while maintaining comprehensive label coverage.
3Reliability
If label interdependence is leveraged, then classification accuracy improves, but handling correlated relationships becomes more complex
Solution Approach 1:
The patent introduces dynamics into the classification chain by allowing the input to each binary classifier to be dynamically adjusted based on predictions from previous classifiers. This dynamic information flow captures label interdependencies and correlated relationships, improving classification accuracy while maintaining a relatively simple binary classifier structure at each stage.
Data Source
AI summary
Various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for converting a multilabel classification model into a sequence of a plurality of binary classification models based on a plurality of label subgroups associated with the multilabel classification model, where the label subgroups comprise an optimal subgroup size, the optimal subgroup size is generated by optimizing an optimization measure defined by a subgroup size variable and a total inner group correlation measure, and identifying label membership to a particular subgroup by using a mixed integer linear program model.