NeurIPS 2026 accepted papers by research area
6,080 papers were accepted at NeurIPS 2026 (+268 vs 2025). Here is how they split across 20 research areas, and which topics gained the most ground.
See NeurIPS 2026 in the interactive map
Research areas
| Area | Papers | Share | Change in share vs 2025 |
|---|---|---|---|
| Large language models | 957 | 15.7% | -0.8 pts |
| Architectures and efficiency | 551 | 9.1% | +0.3 pts |
| Probabilistic and scientific modeling | 543 | 8.9% | steady |
| Vision-language and multimodal | 536 | 8.8% | -0.5 pts |
| Image generation | 456 | 7.5% | steady |
| Multi-agent systems and games | 415 | 6.8% | +3.2 pts |
| Reinforcement learning | 395 | 6.5% | +0.9 pts |
| Learning theory and algorithms | 376 | 6.2% | -0.8 pts |
| Label-efficient and robust learning | 349 | 5.7% | -0.2 pts |
| Video | 280 | 4.6% | +0.6 pts |
| 3D vision | 246 | 4.0% | -0.5 pts |
| Graph learning | 175 | 2.9% | -0.6 pts |
| Optimization | 164 | 2.7% | -0.5 pts |
| Bandits and online learning | 126 | 2.1% | -0.2 pts |
| Time series | 122 | 2.0% | +0.3 pts |
| Federated learning and privacy | 112 | 1.8% | -0.2 pts |
| Adversarial robustness and security | 98 | 1.6% | steady |
| Detection and segmentation | 87 | 1.4% | -0.5 pts |
| Continual learning | 59 | 1.0% | steady |
| Point clouds | 33 | 0.5% | -0.3 pts |
Fastest-growing topics at NeurIPS 2026
Change in share of NeurIPS papers from 2025 to 2026, topics with at least 10 papers.
- LLM agents (Multi-agent systems and games): 105 → 312 papers
- AI and society (Large language models): 61 → 113 papers
- Video generation (Video): 92 → 141 papers
- Policy gradient and actor-critic (Reinforcement learning): 55 → 100 papers
- Robot action and navigation (Vision-language and multimodal): 49 → 90 papers
- RL from feedback (Reinforcement learning): 80 → 122 papers
- Variational and generative sampling (Probabilistic and scientific modeling): 68 → 109 papers
- Fast diffusion sampling (Image generation): 168 → 209 papers
- Retrieval and knowledge (Large language models): 55 → 89 papers
- Causal and generative time series (Time series): 23 → 49 papers