Dual-Model Multiclass Classification for Distributed Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multiclass classification methods face challenges in efficiently handling large datasets and distributed learning environments, particularly in scenarios where data is not centralized, requiring robust and adaptive solutions that can learn from new classes and retain existing knowledge.
Innovation Solution
A method involving a dual model approach, comprising a fixed-sized machine learning classification model and an in-class confidence model, is applied in a distributed machine learning system, where training data is divided into full and partial data classes, and each class is assigned to a different node for prediction and confidence value determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data is distributed across multiple nodes in a distributed learning system, then the system can handle large datasets and scale better, but the complexity of training and coordinating multiple nodes increases
Solution Approach 1:
The patent divides the distributed training process into two separate models: a fixed-size ML classification model trained on full data classes, and an in-class confidence model trained on both full and partial data classes. This segmentation allows each model to handle specific aspects of the classification problem independently, reducing the coordination complexity among distributed nodes while maintaining the ability to process large distributed datasets
Solution Approach 2:
The in-class confidence model serves as an intermediary between the distributed data nodes and the final classification decision. It receives inputs from multiple nodes, aggregates confidence values, and produces a unified prediction. This intermediary structure simplifies the coordination requirements by providing a standardized interface for integrating results from distributed training nodes
2Adaptability or versatility
If the classification model is trained to recognize new classes, then the adaptability of the system improves, but the model size and computational resources required increase
Solution Approach 1:
The patent implements a dynamic two-stage training approach where the fixed-size ML classification model handles established classes, while the in-class confidence model dynamically adapts to new classes by training on partial data classes. This dynamic structure allows the system to accommodate new classes without permanently increasing the base model size, as the confidence model can be retrained with new class data while the fixed-size model remains unchanged
Solution Approach 2:
The system uses partial data classes to train the in-class confidence model, focusing computational resources only on the specific classes needed for adaptation rather than retraining the entire system. This partial action approach allows the system to adapt to new classes with reduced computational overhead compared to full model retraining
3Measurement precision
If all training data is used to train a single comprehensive model, then the model achieves high accuracy across all classes, but the training time and computational resources increase significantly
Solution Approach 1:
The training data and training process are segmented into full data classes and partial data classes, with different models trained on different data subsets. The fixed-size ML classification model is trained only on full data classes, while the in-class confidence model is trained on both full and partial data classes. This segmentation allows parallel training of multiple smaller models instead of one large model, reducing overall training time while maintaining comprehensive class coverage
Solution Approach 2:
The fixed-size ML classification model is trained in advance on full data classes to establish baseline classification capabilities. This preliminary action allows the system to quickly handle common classes without requiring the more time-consuming in-class confidence model for every prediction, thus reducing average processing time while maintaining accuracy for the majority of cases
Data Source
AI summary
A method for receiving a full training data set including a plurality of individual training data set, dividing the plurality of individual training sets into N classes, where N is an integer greater than three, dividing the N classes into M full data classes and N-M partial data classes, performing training to obtain a trained fixed size machine learning (ML) classification model and a trained in-class confidence model, outputting a first set of prediction value(s) based on the performance of training, distributing each class of the N classes of individual training data sets to a different node of a distributed machine learning system; and outputting, from the nodes of the distributed machine learning system, a second set of prediction value(s) for each class of the N classes.