Reproduce “Using Synthetic Data to Train Neural Networks is Model-Based Reasoning” 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

Tuan Anh Le, Atilim Gunes Baydin, Robert Zinkov, Frank Wood

The authors connect synthetic-data training to approximate inference by training a neural network to propose latent explanations for observations generated by a probabilistic model. In Captcha experiments, original models exceeded 90% recognition on synthetic data but had practically zero success on real images; broader generators yielded 81% on Wikipedia and 42% on Facebook, showing results bounded by generator coverage.

Summary by alphaXiv.

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