Reproduce “Efficient Training of Neural SDEs Using Stochastic Optimal Control” 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
Rembert Daems, Manfred Opper, Guillaume Crevecoeur, Tolga Birdal
The authors introduce a hybrid method for variational inference in stochastic differential equations, combining a closed-form linear control with a learned nonlinear residual. The analytic control assumes linear drift, state-independent diffusion, and Gaussian observations. On 500 days of Treasury-bill data, it has lower starting negative ELBO and faster convergence than a neural-network baseline for Brownian and Markov-approximated fractional Brownian noise, in one dimension.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.