Reproduce “Controlling Contrastive Self-Supervised Learning with Knowledge-Driven Multiple Hypothesis: Application to Beat Tracking” 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

Antonin Gagnere, Slim Essid, Geoffroy Peeters

The authors present KD-MHL, a self-supervised pretraining method for beat tracking that keeps beat interpretations from pulse peaks, scores them with music-derived evidence, and trains on hypotheses. In eight-fold cross-validation, its pseudo-label self-training variant reports 89.3 versus 88.0 for Hung et al. (2022); the aggregate averages F1, CMLt, and AMLt across 11 cells, but scores lower on RWC Popular.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.