Reproduce “Multi-Dimensional Recurrent Neural Networks” 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
Alex Graves, Santiago Fernandez, Juergen Schmidhuber
The authors introduce multi-dimensional recurrent neural networks (MDRNNs) for image segmentation, passing information along every spatial dimension and combining scans from different corners. They extend long short-term memory (LSTM) with dimension-specific recurrent connections and forget gates. On warped MNIST, MDRNNs achieve 6.8% image error versus 11.3% for a convolutional baseline, but score 1.1% versus 0.9% on clean MNIST; the warped comparison uses an unaugmented baseline.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.