Reproduce “SoftSort: A Continuous Relaxation for the argsort Operator” 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
Sebastian Prillo, Julian Martin Eisenschlos
SoftSort replaces hard sorting with a differentiable weighting rule for learning through ordered items. It sorts the scores, compares each sorted target score with every original score, and applies a row-wise softmax to negative distances. On handwritten-number sorting at sequence length 15, its exact-permutation accuracy matches NeuralSort; timing tests show faster execution, including after NeuralSort’s implementation is improved.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.