Categorization with Interdependency Reweighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Double-view categorization models become complex and inefficient when interdependencies between categorization dimensions are not considered, leading to redundant models and poor performance in tasks that require categorization across multiple dimensions.
Innovation Solution
A method and apparatus that reweight probability values from one categorization dimension based on the probability values of another dimension, allowing for interdependency adjustments without constructing a new combination model, using reweighting or direct approaches to derive categories that account for interdependencies between dimensions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a combined categorization model is constructed to account for interdependencies between dimensions, then categorization accuracy is improved, but model complexity increases significantly
Solution Approach 1:
The patent segments the complex combined categorization model into separate component models for each categorization dimension. Each dimension is modeled independently, and the final categorization is obtained by combining results from these segmented models, thereby reducing overall model complexity while maintaining the ability to capture interdependencies.
Solution Approach 2:
The patent introduces probability values as intermediary elements that mediate between component models and final categorization decisions. These probability values represent the likelihood of objects belonging to categories in different dimensions and serve as intermediaries to combine results from independent models without requiring a complex integrated model.
2Productivity
If separate categorization models are trained for each dimension, then training efficiency is improved, but model redundancy increases
Solution Approach 1:
The patent makes the component categorization models universal by designing them to serve multiple functions. Each component model not only categorizes objects in its specific dimension but also provides probability values that are reused in combining results across dimensions, eliminating redundancy while maintaining training efficiency.
Solution Approach 2:
The patent merges the functionality of separate categorization models by combining their probability value outputs to achieve multi-dimensional categorization. This merging occurs at the probability level rather than requiring a fully integrated model, thus reducing redundancy while preserving the efficiency benefits of separate training.
3Loss of time
If existing component models are reused for combined categorization, then development time is reduced, but adaptability to interdependencies is limited
Solution Approach 1:
The patent introduces dynamics into the system by allowing probability values from component models to be adjusted and combined flexibly based on interdependencies between dimensions. This dynamic approach enables existing models to adapt to new categorization requirements without retraining, as the probability combination mechanism can be configured to reflect different interdependency relationships.
Data Source
AI summary
In categorizing an object respective to at least two categorization dimensions each defined by a plurality of categories, a probability value indicative of the object is determined for each category of each categorization dimension. A categorization label for the object is selected respective to each categorization dimension based on (i) the determined probability values of the categories of that categorization dimension and (ii) the determined probability values of categories of at least one other of the at least two categorization dimensions.