Classification Accuracy via Probability Vector Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Typical classifiers often inaccurately classify inputs as out-of-domain (OOD) when the maximum probability value is below a threshold, leading to decreased classification accuracy, even if the input actually belongs to the class with the maximum value.
Innovation Solution
A processor-implemented classification method that determines a first probability vector, identifies confusing classes, adjusts the probabilities based on a ratio, and compares the adjusted vector with a second probability vector to improve classification accuracy by avoiding incorrect OOD determinations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a threshold-based classification method is used to determine in-domain or out-of-domain inputs, then the classification process is simple and fast, but the classification accuracy decreases when the maximum probability is below the threshold
Solution Approach 1:
The patent applies preliminary action by performing multiple classifications using different classification models before making a final determination. Instead of relying on a single threshold-based classification, the system conducts preliminary classifications with multiple models to generate probability vectors, then uses these results to make a more accurate final classification decision, thereby improving accuracy without significantly compromising speed.
Solution Approach 2:
The patent introduces an intermediary mechanism by using multiple classification models to generate probability vectors as intermediate results. These probability vectors serve as mediators between the raw input and the final classification decision, allowing the system to evaluate confidence levels and make more accurate determinations about whether inputs are in-domain or out-of-domain.
2Device complexity
If a single classification model is used, then the device complexity is low, but the classification accuracy is insufficient for ambiguous inputs
Solution Approach 1:
The patent applies segmentation by dividing the classification task into multiple independent classification models, each specializing in different aspects of the classification problem. Instead of using one complex model, the system segments the work into multiple simpler models that process the input independently, then combines their probability vectors to achieve higher overall accuracy while maintaining manageable complexity.
Solution Approach 2:
The patent merges the results from multiple classification models by combining their probability vectors through comparison operations. The system merges the strengths of different models by evaluating their respective confidence levels and using the combined information to make a final classification decision, thereby achieving higher accuracy than any single model could provide alone.
Data Source
AI summary
A processor-implemented classification method includes: determining a first probability vector including a first probability, for each of a plurality of classes, resulting from a classification of an input with respect to the classes; determining, based on the determined first probability vector, whether one or more of the classes represented in the first probability vector are confusing classes; adjusting, in response to one or more of the classes being the confusing classes, the determined first probability vector based on a first probability of each of the confusing classes and a maximum value of the first probabilities; determining a second probability vector including a second probability, for each of the classes, resulting from another classification of the input with respect to the classes; and performing classification on the input based on a result of a comparison between the determined second probability vector and the adjusted first probability vector.