Research map / Image generation / Text-to-image diffusion
Kernelized Bayesian Softmax for Text Generation
Ning Miao, Hao Zhou, Chengqi Zhao, Wenxian Shi, Lei Li
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 |
|---|---|---|---|
| replace softmax with KerBS | Seq2Seq · IWSLT'16 De->En MT | BLEU-4: 25.91 -> 27.28 | Improved |
| replace softmax with KerBS | Transformer · IWSLT'16 De->En MT | BLEU-4: 29.61 -> 30.90 | Improved |
| replace softmax with MoS | Seq2Seq · IWSLT'16 De->En MT | BLEU-4: 25.91 -> 26.45 | Improved |
| replace softmax with MoS | Transformer · IWSLT'16 De->En MT | BLEU-4: 29.61 -> 28.54 | Hurt |
| replace softmax with KerBS | GRU LM · One-Billion-Word subset LM | PPL: 103.12 -> 102.17 | Improved |
| replace softmax with KerBS | Seq2Seq · DailyDialog | BLEU-1: 16.56 -> 17.85 | Improved |
| replace softmax with KerBS | Transformer · DailyDialog | BLEU-1: 10.61 -> 10.90 | Improved |
| replace softmax with KerBS | Seq2Seq · DailyDialog human eval | Human Eval.: 1.24 -> 1.40 | Improved |
| replace softmax with MoS | Seq2Seq · DailyDialog | BLEU-1: 16.56 -> 13.73 | Hurt |
| remove kernel function from KerBS (ablation) | KerBS Seq2Seq · IWSLT'16 De->En MT | BLEU-4: 27.28 -> 26.79 | Hurt |
| use single sense instead of multi-sense (ablation) | KerBS Seq2Seq · IWSLT'16 De->En MT | BLEU-4: 27.28 -> 26.80 | Hurt |
| remove dynamic sense allocation (fixed senses per word) (ablation) | KerBS Seq2Seq · IWSLT'16 De->En MT | BLEU-4: 27.28 -> 27.00 | Hurt |
Closest papers by meaning
- A Contrastive Framework for Neural Text Generation (NeurIPS 2022 · similarity 0.86)
- Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe (NeurIPS 2024 · similarity 0.84)
- Improving Open-Ended Text Generation via Adaptive Decoding (ICML 2024 · similarity 0.83)
- Compressed and Smooth Latent Space for Text Diffusion Modeling (NeurIPS 2025 · similarity 0.83)
- Reflection-Window Decoding: Text Generation with Selective Refinement (ICML 2025 · similarity 0.83)
- Decoding Game: On Minimax Optimality of Heuristic Text Generation Strategies (ICLR 2025 · similarity 0.83)
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.
