Reproduce “Rapid Image Labeling via Neuro-Symbolic Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yifeng Wang, Zhi Tu, Yiwen Xiang, Shiyuan Zhou, Xiyuan Chen, Bingxuan Li, Tianyi Zhang
The research introduces "Rapid," a neuro-symbolic framework that automates image labeling by inferring human-readable logic rules from minimal expert-annotated data. It achieves an average accuracy improvement of 11.75% to 12.03% over baselines in specialized domains like Glaucoma Diagnosis and Bird Species Labeling, reaching peak performance within 13 active learning iterations.
Key insights
- Synergistically combining neural networks for perception with symbolic AI for reasoning enables efficient learning from small datasets and yields interpretable, editable labeling rules.
- Direct expert rule editing is crucial for rapidly embedding domain knowledge and significantly accelerating learning and accuracy, especially in complex specialized domains.
- A multi-criteria active learning strategy, balancing conflict-based informativeness with diversity, effectively minimizes human annotation effort while refining rules.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.