Reproduce “Hierarchical Multi-Positive Contrastive Learning for Patent Image Retrieval” 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
Kshitij Kavimandan, Angelos Nalmpantis, Emma Beauxis-Aussalet, Robert-Jan Sips
A Hierarchical Multi-Positive Contrastive Learning (HMCL) approach enhances patent image retrieval by explicitly incorporating hierarchical classification (e.g., Locarno International Classification) into the learning process. This method improves retrieval performance across various architectural scales and hierarchical granularities while being effective with low-parameter models.
Key insights
- Explicitly leveraging pre-existing hierarchical patent classifications (like Locarno International Classification) significantly improves the semantic organization of learned image embeddings.
- The multi-positive weighting in contrastive loss effectively guides models to learn representations that reflect varying degrees of similarity across hierarchical levels.
- The proposed HMCL approach is particularly beneficial for computationally efficient, lower-parameter models, making it practical for real-world deployment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.