scientific-semi-supervised-learning

What problem does it solve?

This Skill provides end-to-end semi-supervised learning pipelines that leverage small amounts of labeled data along with large pools of unlabeled data, enabling model improvement through self-training, label propagation, and pseudo-labeling.

Core Features & Use Cases

• Self-Training: iteratively expand labeled data by training on confident predictions from unlabeled samples.
• Label Propagation: graph-based spreading of labels to unlabeled data to improve class coverage.
• Pseudo-Labeling Quality Evaluation: assess reliability of generated pseudo-labels and tune thresholds for safer labeling.
• ToolUniverse integration: OpenML benchmarking support for standardized evaluation on external science datasets.

Quick Start

Run a semi-supervised learning workflow by providing your labeled and unlabeled data to iteratively label and improve the model.