Reproduce “PROLA: Principal-Orthogonal Low-rank Adaptation for Predictive Spatiotemporal Weather Downscaling” 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
Minseo Yoon, Minseong Bae, Sojin Lee, Hyunwoo J. Kim
PROLA improves low-rank adaptation for predictive spatiotemporal weather downscaling by dividing a fixed rank budget between pretrained-direction updates and updates projected outside those directions, then scaling rank-one updates using gradient signals. On WeatherBench 2, its 6h Z500 RMSE is 72.35 ± 0.64, versus 77.51 ± 0.94 for AdaLoRA. This comparison does not establish faster training or universal gains over full fine-tuning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.