Reproduce “Less is More: Parameter-Free Text Classification with Gzip” 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
Zhiying Jiang, Matthew Y. R. Yang, Mikhail Tsirlin, Raphael Tang, Jimmy Lin
The authors use gzip-based normalized compression distance and a two-nearest-neighbor rule to classify new texts by borrowing labels from the most compressibly similar labeled examples, without fitting or pretrained models. Gzip is competitive with non-pretrained neural methods on six of seven in-distribution datasets and beats BERT and multilingual BERT on five selected Roman-script out-of-distribution datasets, not arbitrary shifts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.