Reproduce “Modelling Non-Smooth Signals with Complex Spectral Structure” 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
Wessel P. Bruinsma, Martin Tegnér, Richard E. Turner
The authors introduce rough Gaussian process convolution models that retain a learned spectrum while allowing less smooth time-series signals. RGPCM uses an Ornstein–Uhlenbeck input and a white-noise filter prior; structured Gibbs inference preserves dependence between latent variables. On NASDAQ crude-oil prices, RGPCM reports mean log loss of 1.681 ± 0.061, significantly lower than GPCM. This task-specific comparison does not isolate which prior change caused improvement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.