Research map / Architectures and efficiency / RNNs and normalization
Differentiable Sorting Networks for Scalable Sorting and Ranking Supervision
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 the sorting operators with differentiable odd-even sorting networks | four-digit MNIST ranking supervision · four-digit MNIST, n=5 | EM: 71.5 -> 86.3 | Helped |
| Use the bitonic sorting network rather than odd-even | SVHN ranking supervision · SVHN, n=32 | EW: 36.6 -> 42.4 | Helped |
| Add the activation replacement trick (ablation) | differentiable sorting networks · MNIST, n=32 | EW: 61.5 -> 69.1; 62.8 -> 67.3 | Helped |
| Add the activation replacement trick (ablation) | differentiable sorting networks · SVHN, n=32 | EW: 28.5 -> 36.6; 35.0 -> 42.4 | Helped |
| Increase ART intensity from lambda=0.25 to lambda=0.4 (ablation) | Bitonic sorting network · MNIST, n=32 | EM5: 78.2% to 80.97% | Helped |
| Train on larger ordered sets | Bitonic sorting network · MNIST, up to n=1024 | EM5: 78.20; 82.80 | Helped |
| Use differentiable top-k supervision instead of Softmax Cross-Entropy | CIFAR-10, Vanilla CNN · CIFAR-10 | accuracy: 87.2% -> 88.0% | Helped |
| Use differentiable top-k supervision instead of Softmax Cross-Entropy | CIFAR-10, ResNet18 · CIFAR-10 | accuracy: 91.0% -> 90.9% | No clear effect |
| Use differentiable top-k supervision instead of Softmax Cross-Entropy | CIFAR-100, Vanilla CNN · CIFAR-100 | accuracy: 58.2% -> 56.3% | Hurt |
| Use differentiable top-k supervision instead of Softmax Cross-Entropy | CIFAR-100, ResNet18 · CIFAR-100 | accuracy: 61.9% -> 63.3% | Helped |
| Increase ART intensity beyond lambda=0.5 (ablation) | differentiable sorting networks · varying ART intensities | accuracy | Hurt |
Closest papers by meaning
- Monotonic Differentiable Sorting Networks (ICLR 2022 · similarity 0.92)
- Generalized Neural Sorting Networks with Error-Free Differentiable Swap Functions (ICLR 2024 · similarity 0.91)
- Differentiable Ranking and Sorting using Optimal Transport (NeurIPS 2019 · similarity 0.89)
- Stochastic Optimization of Sorting Networks via Continuous Relaxations (ICLR 2019 · similarity 0.89)
- LapSum - One Method to Differentiate Them All: Ranking, Sorting and Top-k Selection (ICML 2025 · similarity 0.85)
- Training Binary Neural Networks via Gaussian Variational Inference and Low-Rank Semidefinite Programming (NeurIPS 2024 · 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.