Reproduce “Neural Scaling Laws of Transformer Operator Network” 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
Haoran Yan, Zhongjie Shi, Yuanzhe Xi, Peng Chen, Wenjing Liao
The authors develop a Transformer operator learner mapping input functions to outputs at requested locations. Attention combines representative inputs and output locations; theory bounds error under regularity and input-complexity assumptions. For KdV, fitted test-MSE decay exponents are 2.10 at input-family dimension 4 and 1.47 at dimension 8. The same ordering appears in Burgers’, but these exponents do not estimate the theoretical rates.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.