Reproduce “MahNMF: Manhattan Non-negative Matrix Factorization” 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
Naiyang Guan, Dacheng Tao, Zhigang Luo, John Shawe-Taylor
MahNMF replaces squared-error fitting in non-negative low-rank factorization with the sum of absolute residuals, targeting heavy-tailed corruption. Its rank-one residual iteration offers closed-form updates for small problems, while a smoothed accelerated optimizer is intended for larger ones. On Yale B, it performs better under Laplace and salt-and-pepper corruption, but poorly under Gaussian noise and comparably to KLNMF under Poisson.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.