Codes for work on few-shot learning on chest x-ray images (paper).
Check our website for a brief summary of the paper.
tl;dr : We propose a computationally efficient few-shot learning method for diagnosing chest X-rays, which uses an ensemble of random subspaces and a novel loss function to create well-separated clusters of training data in discriminative subspaces. Our method is almost 1.8 times faster than the popular t-SVD method for subspace decomposition and yields promising results on large-scale CXR datasets.
