Research map / Probabilistic and scientific modeling / Molecules and proteins
Learning Sparse Prototypes for Text Generation
Junxian He, Taylor Berg-Kirkpatrick, Graham Neubig
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 |
|---|---|---|---|
| Use the learned sparse prototype model instead of the NLM baseline | Language modeling · MSCOCO | PPL: 20.0 -> 18.6 | Improved |
| Use the learned sparse prototype model instead of the NLM baseline | Language modeling · Yelp Medium | PPL: 74.7 -> 61.9 | Improved |
| Use the learned sparse prototype model instead of the NLM baseline | Language modeling · Yelp Large | PPL: 34.2 -> 30.2 | Improved |
| Interpolate the sparse model with an NLM | Sparse Neural Editor · Yelp Large | PPL: 30.2 -> 20.2 | Improved |
| Use learned prototypes instead of heuristic pre-clustering | Neural Editor with 778 prototypes · MSCOCO | PPL: 19.5 -> 18.6 | Improved |
| Use learned prototypes instead of heuristic pre-clustering | Neural Editor with 778 prototypes · MSCOCO | POS-BLEU: 40.4 -> 42.5 | Improved |
| Increase Dirichlet concentration α from 10⁻³ to 0.1 (ablation) | Sparse Neural Editor · MSCOCO | PPL: 18.9 -> 18.6 | Improved |
| Increase Dirichlet concentration α from 0.1 to 0.2 (ablation) | Sparse Neural Editor · MSCOCO | PPL: 18.6 -> 19.0 | Hurt |
| Increase Dirichlet concentration α from 0.2 to 0.3 (ablation) | Sparse Neural Editor · MSCOCO | # prototypes: 16K (40%) -> 22K (56%) | Mixed |
| Increase Dirichlet concentration α from 10⁻³ to 0.5 (ablation) | Sparse Neural Editor · Yelp Medium | PPL: 63.6 -> 61.9 | Improved |
| Increase Dirichlet concentration α from 0.5 to 10 (ablation) | Sparse Neural Editor · Yelp Medium | PPL: 61.9 -> 63.2 | Hurt |
| Increase Dirichlet concentration α from 0.7 to 10 (ablation) | Sparse Neural Editor · Yelp Large | PPL: 30.2 -> 30.3 | Hurt |
Closest papers by meaning
- Accelerating Inference of Retrieval-Augmented Generation via Sparse Context Selection (ICLR 2025 · similarity 0.87)
- Retrieval is Accurate Generation (ICLR 2024 · similarity 0.86)
- Parameterized Synthetic Text Generation with SimpleStories (NeurIPS 2025 · similarity 0.86)
- Syntactic and Semantic Control of Large Language Models via Sequential Monte Carlo (ICLR 2025 · similarity 0.86)
- Guaranteed Generation from Large Language Models (ICLR 2025 · similarity 0.86)
- Compressed and Smooth Latent Space for Text Diffusion Modeling (NeurIPS 2025 · similarity 0.86)
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.
