Research map / Learning theory and algorithms / Optimal transport
Differentiable Top-k with Optimal Transport
Yujia Xie, Hanjun Dai, Minshuo Chen, Bo Dai, Tuo Zhao, Hongyuan Zha, Wei Wei, Tomas Pfister
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add SOFT top-k beam search training to LSTM seq2seq | Bahdanau et al. LSTM · WMT14 En-Fr | BLEU: 35.38 -> 36.27 | Improved |
| use SOFT top-k gradient instead of vanilla top-k with ignored gradient in beam search training (ablation) | Beam search training · WMT14 En-Fr | BLEU: 35.84 -> 36.27 | Improved |
| incorporate beam search into training (even with vanilla top-k) vs teacher forcing only (ablation) | LSTM seq2seq · WMT14 En-Fr | BLEU: 35.38 -> 35.84 | Improved |
| train kNN end-to-end with SOFT top-k instead of two-stage pretrained CNN + kNN | kNN classifier · CIFAR-10 | accuracy: 91.1% -> 92.6% | Improved |
| SOFT top-k instead of k-times softmax relaxation | kNN classifier · CIFAR-10 | accuracy: 92.2% -> 92.6% | Improved |
| SOFT top-k instead of NeuralSort | kNN classifier · MNIST | accuracy: 99.5% -> 99.4% | No clear effect |
Closest papers by meaning
- ST$_k$: A Scalable Module for Solving Top-k Problems (NeurIPS 2024 · similarity 0.85)
- LapSum - One Method to Differentiate Them All: Ranking, Sorting and Top-k Selection (ICML 2025 · similarity 0.85)
- Statistical Perspective of Top-K Sparse Softmax Gating Mixture of Experts (ICLR 2024 · similarity 0.82)
- Low-Rank Optimal Transport through Factor Relaxation with Latent Coupling (NeurIPS 2024 · similarity 0.82)
- Faster Differentially Private Top-$k$ Selection: A Joint Exponential Mechanism with Pruning (NeurIPS 2024 · similarity 0.81)
- Double-Bounded Optimal Transport for Advanced Clustering and Classification (AAAI 2024 · similarity 0.81)
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.
