Expert-Guided Label Propagation for Semi-Supervised Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The laborious and costly process of data collection and labeling for machine learning models, particularly the scarcity of labeled samples compared to abundant unlabeled samples, hinders the efficiency of current machine learning pipelines.
Innovation Solution
A method for label propagation that involves receiving a labeled dataset, computing latent representation spaces, generating a k-nearest neighbor similarity graph, augmenting it with an expert-derived similarity graph, and propagating labels to unlabeled samples using a combined similarity graph, allowing for semi-supervised learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If manual labeling is used to label unlabeled samples, then the number of labeled samples increases, but the cost and time consumption increase significantly
Solution Approach 1:
The system uses automatically generated pseudo-labels from the pre-trained model to label unlabeled samples, enabling the data labeling process to serve itself without requiring manual human intervention for every sample
Solution Approach 2:
The system performs pre-training on initially labeled samples before using the pre-trained model to generate pseudo-labels for unlabeled samples, preparing the model in advance to perform the labeling task efficiently
2Quantity of substance
If manual labeling is used to label unlabeled samples, then the number of labeled samples increases, but the cost increases significantly
Solution Approach 1:
The system uses automatically generated pseudo-labels from the pre-trained model to label unlabeled samples, enabling the data labeling process to serve itself without requiring manual human intervention for every sample
Solution Approach 2:
The system creates copies of the labeling pattern learned from labeled samples and applies them to unlabeled samples through pseudo-labeling, avoiding the need for expensive manual copying of labels
3Reliability
If only labeled samples are used for training, then the model training is straightforward, but the performance is limited due to scarcity of labeled data
Solution Approach 1:
The system merges labeled samples with unlabeled samples that have been assigned pseudo-labels to form a combined training dataset, leveraging both supervised and unsupervised learning signals to improve model performance
Solution Approach 2:
The system transitions from a single-dimension supervised learning approach to a multi-dimension semi-supervised learning approach by incorporating unlabeled data with pseudo-labels as an additional training dimension
Data Source
AI summary
A method includes receiving a labeled dataset that includes a plurality of labeled samples and initially training the machine learning model using the labeled dataset. The method also includes receiving an unlabeled dataset that includes a plurality of unlabeled samples. The method also includes computing latent representation spaces for each respective sample of the plurality of labeled samples and each respective sample of the plurality of the unlabeled samples. The method also includes generating a k-nearest neighbor similarity graph based on the latent representation spaces, generating a combined similarity graph by augmenting the k-nearest neighbor similarity graph using an expert-derived similarity graph, and propagating, using the combined similarity graph, labels to each respective sample of the plurality of unlabeled samples. The method also includes subsequently training the machine learning model using the labeled dataset and the unlabeled dataset having samples propagated with labels using the combined similarity graph.