Reproduce “Masked Autoencoders for Ultrasound Signals: Robust Representation Learning for Downstream Applications” 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
Immanuel Roßteutscher, Klaus S. Drese, Thorsten Uphues
The authors adapt masked-autoencoder pre-training to one-dimensional ultrasound signals by masking patches and reconstructing them. They retain the encoder to classify the onset of a tone burst. For M-dh64, synthetic pre-training yielded 87.46% ± 0.30% top-1 accuracy, versus 78.68% ± 3.11% after measured-data pre-training. This comparison uses 60K versus 12K pre-training examples and does not isolate data source or variability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.