Reproduce “Robust Unsupervised Learning via L-Statistic Minimization” 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
Andreas Maurer, Daniela A. Parletta, Andrea Paudice, Massimiliano Pontil
The authors introduce rank-weighted fitting for k-means and principal subspace analysis, ranking each candidate’s reconstruction losses so smaller losses receive greater weight. Outlier status is therefore model-dependent. Under mixture and low-loss-concentration conditions, the fit is within δ of a reference model’s target error. On contaminated Fashion-MNIST k-means, RKM’s error is 25.98 versus 34.39 for K-MEANS++; this is not a guarantee for arbitrary contamination.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.