Classifier Training Using Noisy Samples via Weighted Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training classifiers using highly noisy datasets is challenging due to the high false positive rates, making it difficult to annotate and label training data effectively, especially when the noise level exceeds 50%.
Innovation Solution
A system that clusters input data using weighted deep embedded clustering (W-DEC) and computes potential scores for each cluster, iteratively refining cluster quality, and trains classifiers by sampling negative and positive datasets uniformly and non-uniformly based on these scores, respectively, to reduce noise levels and improve classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training data is annotated and labeled manually to reduce noise, then classification accuracy is improved, but time consumption and resource costs increase significantly
Solution Approach 1:
The system performs self-labeling by automatically identifying and labeling positive samples through clustering and potential score computation, eliminating the need for manual annotation of all training data while maintaining high classification accuracy
Solution Approach 2:
The system pre-processes the noisy positive data by clustering samples and computing potential scores before training the classifier, thereby preparing clean labeled data in advance and avoiding time-consuming manual annotation during the training phase
2Measurement precision
If uniform sampling is used for both positive and negative datasets, then training simplicity is maintained, but classification accuracy deteriorates due to noise in positive samples
Solution Approach 1:
The system applies different sampling strategies to different datasets: uniform sampling for clean negative data and non-uniform potential-score-based sampling for noisy positive data, optimizing the training process for each data type's specific characteristics
Solution Approach 2:
The system changes the sampling parameter from uniform distribution to non-uniform distribution based on potential scores for positive samples, allowing higher-weighted sampling of cleaner positive samples and lower-weighted sampling of noisier ones
Data Source
AI summary
An example system includes a processor to receive input data comprising noisy positive data and clean negative data. The processor is to cluster the input data. The processor is to compute a potential score for each cluster of the clustered input data. The processor is to iteratively refine cluster quality of the clusters using the potential scores of the clusters as weights. The processor is to train a classifier by sampling the negative dataset uniformly and the positive set in a non-uniform manner based on the potential score.