Reproduce “Learning from Context: Leave-One-Out Prediction in Jump Diffusions” 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

Christian Wald

The authors develop leave-one-out prediction for embedding-space jump diffusion, predicting each target from other corrupted positions. They use these contextual predictions for reset destinations while retaining the full posterior for continuous drift, preserving ideal process marginals under stated assumptions. On LM1B, RJD-LOO reaches best MAUVE 0.9039 ± 0.0058 versus RJD’s 0.2019 ± 0.0032 across separately tuned decoding settings; raw LOO drift lacks that guarantee.

Summary by alphaXiv.

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