Convex Optimized Stochastic Vector Sampling for ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Supervised learning models require large volumes of labeled data for accurate multiclass classification, which is tedious and time-consuming, especially in cases of class imbalance where data distribution is uneven, leading to inaccurate predictive performance.
Innovation Solution
The system uses intelligently selected multiclass vectors by applying t-SNE for dimensionality reduction, centroid-based clustering, and convex optimization to identify optimal clusters, selecting representative vectors for labeling, and performing inductive learning and forgetting to improve model performance without extensive data labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If large volumes of labeled data are used for training supervised learning models, then prediction accuracy is improved, but labeling effort and time consumption increase significantly
Solution Approach 1:
The system performs preliminary actions by automatically selecting representative vectors from unlabeled data using t-SNE dimensionality reduction and convex optimization before labeling is needed. This pre-processing identifies the most informative data points that would yield highest training accuracy with minimal labeling effort, thereby resolving the contradiction between needing large labeled datasets and the time-consuming labeling process.
Solution Approach 2:
The invention extracts only the essential representative vectors from the large pool of unlabeled data using automated selection algorithms. By taking out just the most informative samples (approximately 10-100 representative vectors per class) rather than labeling all data, the system achieves high prediction accuracy while dramatically reducing labeling time and effort.
2Measurement precision
If more data is collected for multiclass classification, then model performance is improved, but class imbalance problems are exacerbated
Solution Approach 1:
The system applies local quality by treating different classes differently during the representative vector selection process. Using convex optimization, the algorithm identifies representative vectors that are most informative for distinguishing between classes, automatically adjusting the selection criteria to account for class imbalance. This ensures that classes with fewer available data points still receive appropriate representation in the final training set.
Solution Approach 2:
The invention changes the parameter of data representation by transforming the original high-dimensional feature space through t-SNE dimensionality reduction. This transformation creates a new representation space where classes are better separated and more representative vectors can be identified, allowing the system to achieve good model performance with fewer data points and reducing the impact of class imbalance.
3Measurement precision
If manual labeling of large datasets is performed, then training data quality is improved, but productivity decreases
Solution Approach 1:
The system performs self-service by automatically selecting representative vectors and preparing training data without requiring manual intervention. The automated pipeline uses t-SNE for dimensionality reduction, convex optimization for cluster identification, and representative vector selection algorithms to create high-quality training datasets independently, thereby maintaining data quality while dramatically improving processing efficiency and productivity.
Solution Approach 2:
The invention replaces the mechanical manual labeling process with automated computational algorithms. Instead of human annotators manually reviewing and labeling data points, the system uses machine-based algorithms (t-SNE, convex optimization, clustering) to automatically identify and select representative vectors, substituting mechanical manual labor with automated intelligent selection that achieves both high quality and high productivity.
Data Source
AI summary
Systems and methods are described for training a machine learning model using intelligently selected multiclass vectors. According to an embodiment, a processing resource of a computing system receives a first set of un-labeled feature vectors. The first set feature vectors are homomorphically translated using a T-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm to obtain a second set of feature vectors with reduced dimensionality. The second set of feature vectors are clustered to obtain an initial set of clusters using centroid-based clustering. An optimal set of clusters is identified among the initial set of clusters by performing a convex optimization process on the initial set of clusters. For each cluster of the optimal set of clusters, a representative vector from the cluster is selected for labeling.