Reproduce “Learning Fractals by Gradient Descent” 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

Cheng-Hao Tu, Hong-You Chen, David Carlyn, Wei-Lun Chao

The authors develop a differentiable pipeline to learn iterated function system rules that make rendered fractals resemble target images. It optimizes affine transformations through image loss and renders point sequences into pixels. On MNIST, the best setting reports mean squared error of 0.0202 ± 0.0005, versus 0.1724 ± 0.0000 for an inverse-encoder baseline; evaluation selects the lowest error among 100 generated images. The grayscale method struggles with detail.

Summary by alphaXiv.

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