Research map / Probabilistic and scientific modeling / Variational and generative sampling
SoftSort: A Continuous Relaxation for the argsort Operator
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace NeuralSort with SoftSort | Differentiable sorting for large-MNIST permutation prediction · Large-MNIST sorting, n=3, 5, 7, 9, 15 | proportion of correct permutations: 0.921 ± 0.006 -> 0.918 ± 0.005 (n=3); 0.797 ± 0.010 -> 0.796 ± 0.009 (n=5); 0.663 ± 0.016 -> 0.666 ± 0.016 (n=7); 0.547 | No clear effect |
| Replace NeuralSort with SoftSort | Differentiable quantile regression · Large-MNIST, n=5, 9, 15 | mean squared error: 21.52 -> 23.44 (n=5); 15.00 -> 19.26 (n=9); 18.81 -> 15.54 (n=15) | Mixed |
| Replace NeuralSort with SoftSort in differentiable kNN | kNN classifier · MNIST | test classification accuracy: 99.5% -> 99.37% | Hurt |
| Replace NeuralSort with SoftSort in differentiable kNN | kNN classifier · Fashion-MNIST | test classification accuracy: 93.5% -> 93.60 % | Helped |
| Replace NeuralSort with SoftSort in differentiable kNN | kNN classifier · CIFAR-10 | test classification accuracy: 90.7% -> 92.03% | Helped |
| Use SoftSort instead of NeuralSort | Sorting operator implementation · Speed comparison, CPU and GPU | runtime: 80% slower on CPU; 40% slower on GPU | Helped |
| Increase the learning rate and temperature used for NeuralSort sorting experiments | NeuralSort · Large-MNIST sorting | proportion of correct permutations | Helped |
| Use absolute distance instead of squared distance in SoftSort (ablation) | SoftSort · Synthetic speed-comparison task | convergence speed | Hurt |
| Use absolute distance instead of squared distance in SoftSort (ablation) | SoftSort · Operator choice during experimentation | performance | Helped |
| Increase k in differentiable kNN (ablation) | SoftSort-based kNN · MNIST, Fashion-MNIST, and CIFAR-10 | classification performance | Helped |
| Use cross-entropy loss instead of negative predicted class probability (ablation) | Differentiable kNN with SoftSort or NeuralSort · MNIST, Fashion-MNIST, and CIFAR-10 | classification performance | Hurt |
Closest papers by meaning
- Stochastic Optimization of Sorting Networks via Continuous Relaxations (ICLR 2019 · similarity 0.89)
- Generalized Neural Sorting Networks with Error-Free Differentiable Swap Functions (ICLR 2024 · similarity 0.84)
- Incremental Gradient Descent with Small Epoch Counts is Surprisingly Slow on Ill-Conditioned Problems (ICML 2025 · similarity 0.83)
- OT4P: Unlocking Effective Orthogonal Group Path for Permutation Relaxation (NeurIPS 2024 · similarity 0.83)
- Generalized Gradient Norm Clipping & Non-Euclidean $(L_0,L_1)$-Smoothness (NeurIPS 2025 · similarity 0.82)
- LapSum - One Method to Differentiate Them All: Ranking, Sorting and Top-k Selection (ICML 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.