Graph-Based Sub-Sampling for Accelerated Active Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Active learning in machine learning is computationally burdensome due to the need to scan large unlabeled data pools for uncertainty and diversity, leading to impractical processing times on large datasets.
Innovation Solution
The use of a proximity graph generated from a Variational Auto Encoder's latent space for diffusing label knowledge and selecting the top k most uncertain nodes, reducing the dataset size to be processed by focusing on nodes with high uncertainty values, thereby accelerating the active learning process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If active learning scans large unlabeled data pools to identify uncertain and diverse samples, then the quality of selected samples for training is improved, but the computational time and resources required become prohibitively large
Solution Approach 1:
The patent segments the large unlabeled data pool into smaller subsets using clustering algorithms. Instead of scanning the entire dataset, the system divides it into manageable clusters and selectively processes representative samples from each cluster, thereby maintaining sample selection quality while dramatically reducing computational time and resources.
2Reliability
If active learning processes the entire unlabeled data pool to ensure comprehensive uncertainty evaluation, then the reliability of model improvement is enhanced, but the computational resources required become impractical
Solution Approach 1:
The patent extracts a representative subset of unlabeled data from the complete data pool by identifying and selecting diverse, uncertain samples through clustering and uncertainty estimation. This extraction approach maintains the reliability of model improvement by ensuring the selected subset adequately represents the full data distribution, while consuming far fewer computational resources than processing the entire pool.
3Productivity
If the dataset is reduced to a smaller subset for processing, then the computational efficiency is improved, but the risk of missing important diverse samples increases
Solution Approach 1:
The patent performs preliminary clustering and diversity analysis on the unlabeled data pool before final sample selection. By pre-organizing data into clusters and identifying representative samples from each cluster in advance, the system ensures comprehensive diversity coverage in the reduced subset, thereby maintaining adaptability and versatility while achieving high computational efficiency during the main active learning process.
Data Source
AI summary
In some embodiments, there is provided receiving, as an input to a first machine learning model, a plurality of data; learning, by the first machine learning model and based at least on the plurality of data, a latent space; generating, based on the plurality of data and the latent space, a proximity graph, wherein label knowledge from labeled data is diffused on a plurality of nodes of the proximity graph; filtering, by the proximity graph, the plurality of nodes to provide a top k most uncertain nodes, wherein the top k most uncertain nodes form a subset of a plurality of unlabeled data; and providing the subset of the plurality of unlabeled data to a second machine learning model comprised in an active learning process. Related system, methods, and articles of manufacture are also disclosed.