Reproduce “PACE: Pretrained Audio Continual 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
Chang Li, Kanglei Zhou, Liyuan Wang
PACE addresses continual audio learning by selectively adapting pretrained backbones. It uses low-rank adapters, a statistics-based analytic classifier, and projected updates to limit interference across sessions. On EAT-backbone TIMIT-2, PACE achieves 90.95% mean accuracy after each task over accumulated classes, versus RanPAC’s 85.63%; joint training reaches 95.22%. These benchmark results do not establish performance across all audio models or real-world streams.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.