Reproduce “Universal randomised signatures for generative time series modelling” 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

Francesca Biagini, Lukas Gonon, Niklas Walter

The authors introduce a generator-training method that replaces a learned adversarial critic with fixed randomised-signature features, matching average features of real and generated paths. They apply it to unconditional and conditional generation. On held-out S&P 500 returns, NeuralSDE at N = 80 has lower covariance- and autocorrelation-distance scores than its signature-based comparator; at N = 100, its autocorrelation score is worse. These metrics assess selected dependence statistics, not overall realism.

Summary by alphaXiv.

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