Crossbar Array Quantization Step Sizing for Imbalanced Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image object classification models face challenges due to class imbalance, where unequal amounts of training data for different classes lead to degraded classification performance, particularly for classes with insufficient data.
Innovation Solution
A computing device uses a prediction consistency value to classify objects as majority or minority classes based on the consistency of class prediction values from multiple models, then applies appropriate class predictors to improve accuracy, using ensemble models for majority classes and optimized models for minority classes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If classification models are trained with balanced data for all classes, then classification performance for minority classes improves, but training data requirements and training time increase
Solution Approach 1:
The patent applies preliminary action by pre-training classification models with balanced training data before deployment. This preliminary training ensures that minority class categories are properly learned, so when the model encounters imbalanced inference data, it can still accurately classify minority classes without requiring additional training time at inference time.
Solution Approach 2:
The patent segments the classification process into two phases: a training phase with balanced data to learn minority class characteristics, and an inference phase that handles imbalanced data. This segmentation allows the model to acquire knowledge about minority classes during training, then apply that knowledge efficiently during deployment without requiring continuous retraining.
2Measurement precision
If multiple classification models are used to improve accuracy through ensemble methods, then classification performance improves, but device complexity increases
Solution Approach 1:
The patent merges multiple classification models into an ensemble system where each model processes the input image and their predictions are combined. This combining approach leverages the strengths of different models to improve overall classification accuracy, particularly for minority classes, while managing complexity through systematic integration of model outputs.
Solution Approach 2:
The patent implements feedback mechanisms where the predictions from multiple models are aggregated and used to determine final classification results. The system analyzes the consistency and distribution of predictions across models, using this feedback to resolve uncertain cases and improve reliability without requiring excessive computational resources.
Data Source
AI summary
Disclosed is a method that includes generating a prediction consistency value that indicates a consistency of prediction of an object in an input image with respect to class prediction values for the object in an input image from classification models to which the input image is input, and identifying a class of the object. Identifying the class of the object includes, in response to a class type being determined, based on the prediction consistency value, of the object being determined to correspond to a majority class, identifying a class of the object based on a corresponding class prediction value output for the object from a majority class prediction model, and in response to the class type of the object being determined to correspond to a minority class, identifying the class of the object based on another corresponding class prediction value output for the object from a minority class prediction model.