Reproduce “Learning from Limited Labels: Transductive Graph Label Propagation for Indian Music Analysis” 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
Parampreet Singh, Akshay Raina, Sayeedul Islam Sheikh, Vipul Arora
Researchers at IIT Kanpur applied a transductive Label Propagation framework to improve Raga identification and instrument classification in Indian Art Music (IAM). The method effectively generated high-quality pseudo-labels for large, unlabeled audio datasets from the Prasar Bharati Archives, boosting instrument classification accuracy to 91.7% on a gold set and improving Raga classification F1-score by 5% at the file level compared to supervised baselines.
Key insights
- Label Propagation effectively leverages the underlying structure of audio embeddings to propagate label information from a small labeled set to a large unlabeled corpus.
- The iterative refinement process, where classifier embeddings inform graph construction for LP, and LP's pseudo-labels improve classifier training, leads to robust models.
- Semi-supervised learning, specifically LP, offers a scalable and efficient alternative to manual annotation, thereby democratizing data annotation in MIR.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.