Noise-Robust Semi-Supervised Learning Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classification methods are ineffective when dealing with noisy classified observation vectors, as they assume correct classifications for training, which is not always the case due to human errors or other factors, leading to poor performance in classifying unclassified data.
Innovation Solution
A noise-robust distributable semi-supervised learning method is employed, utilizing L1 optimization and sparse coding to suppress noisy classified observation vectors, thereby improving classification accuracy by leveraging a Laplacian matrix decomposition and sparse coefficients update process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing classification methods are used that assume correct classifications for training, then the classification process is simple, but the classification accuracy deteriorates when noisy data is present
Solution Approach 1:
The patent converts the harmful effect of noisy classified observation vectors into a beneficial filtering process. By introducing a noise robust loss function that identifies and suppresses noisy samples, the system transforms data quality issues into an opportunity for improved classification accuracy through selective weighting and outlier detection.
Solution Approach 2:
The patent changes the parameter space by introducing sparse coefficients and noise robust loss functions that modify how training data is weighted. The system adjusts parameters dynamically to identify and suppress noisy samples, transforming the classification approach from static to adaptive parameter management.
2Quantity of substance
If semi-supervised learning is used to label unlabeled data, then the use of labeled data is reduced, but the complexity of the learning process increases
Solution Approach 1:
The patent applies partial action by using only a small portion of labeled data to train the model, while the majority of data remains unlabeled. The semi-supervised learning approach processes only necessary labeled samples intensively, allowing the system to achieve good performance with minimal labeled data input.
Solution Approach 2:
The patent introduces unlabeled data as an intermediary between the limited labeled data and the final classification model. The unlabeled data serves as a bridge that allows the model to learn from the structure and distribution of the full dataset without requiring extensive manual labeling.
3Quantity of substance
If noisy classified observation vectors are used for training, then data availability is maintained, but the performance of classifying unclassified data deteriorates
Solution Approach 1:
The patent converts the presence of noisy data from a harmful factor into a beneficial filtering mechanism. By designing a noise robust loss function that automatically identifies and suppresses noisy samples, the system maintains data availability while improving classification performance through selective data weighting.
Solution Approach 2:
The patent extracts and removes noisy classified observation vectors from the training process through sparse coding and outlier detection. The system identifies problematic samples and excludes or down-weights them, separating the useful signal from the noisy data to improve overall classification performance.
Data Source
AI summary
A computing device predicts occurrence of an event or classifies an object using distributed unlabeled data. A Laplacian matrix is computed using a kernel function. A predefined number of eigenvectors is selected from a decomposed Laplacian matrix to define a decomposition matrix. A gradient value is computed as a function of the defined decomposition matrix, a plurality of sparse coefficients, and a label matrix, a value of each coefficient of the plurality of sparse coefficients is updated based on the computed gradient value, and the computations are repeated until a convergence parameter value indicates the plurality of sparse coefficients have converged. A classification matrix is defined using the plurality of sparse coefficients to determine the target variable value for each observation vector of the plurality of unclassified observation vectors. The target variable value for each observation vector of the plurality of unclassified observation vectors is output.