Reproduce “Learning from Between-class Examples for Deep Sound Recognition” 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
Yuji Tokozume, Yoshitaka Ushiku, Tatsuya Harada
BC learning trains sound-recognition networks on mixtures of recordings from different classes, supervising them to estimate each class’s contribution rather than only predict one label. It uses ratio targets, KL-divergence, and ordinary recordings at test time. On ESC-50, EnvNet-v2 with strong augmentation reached 15.1% ± 0.2 error versus 21.2% ± 0.3, but required twice as many epochs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.