ICLR 2025 accepted papers by research area
3,703 papers were accepted at ICLR 2025 (+1,444 vs 2024). Here is how they split across 20 research areas, and which topics gained the most ground.
See ICLR 2025 in the interactive map
Research areas
| Area | Papers | Share | Change in share vs 2024 |
|---|---|---|---|
| Large language models | 742 | 20.0% | +5.5 pts |
| Architectures and efficiency | 381 | 10.3% | +0.5 pts |
| Image generation | 343 | 9.3% | +1.1 pts |
| Probabilistic and scientific modeling | 288 | 7.8% | -0.9 pts |
| Vision-language and multimodal | 269 | 7.3% | +2.1 pts |
| Label-efficient and robust learning | 230 | 6.2% | -3.6 pts |
| Reinforcement learning | 215 | 5.8% | -2.5 pts |
| Learning theory and algorithms | 175 | 4.7% | -1.0 pts |
| Graph learning | 151 | 4.1% | -1.4 pts |
| Video | 149 | 4.0% | +2.0 pts |
| 3D vision | 133 | 3.6% | +0.3 pts |
| Optimization | 123 | 3.3% | +0.1 pts |
| Multi-agent systems and games | 119 | 3.2% | +0.9 pts |
| Adversarial robustness and security | 97 | 2.6% | -1.0 pts |
| Federated learning and privacy | 64 | 1.7% | -1.0 pts |
| Time series | 64 | 1.7% | -0.6 pts |
| Detection and segmentation | 58 | 1.6% | -0.3 pts |
| Continual learning | 39 | 1.1% | -0.1 pts |
| Bandits and online learning | 38 | 1.0% | steady |
| Point clouds | 25 | 0.7% | -0.1 pts |
Fastest-growing topics at ICLR 2025
Change in share of ICLR papers from 2024 to 2025, topics with at least 10 papers.
- Multimodal LLMs (Vision-language and multimodal): 19 → 132 papers
- Alignment and preferences (Large language models): 24 → 97 papers
- Fast diffusion sampling (Image generation): 58 → 151 papers
- LLM agents (Multi-agent systems and games): 10 → 59 papers
- Reasoning and chain of thought (Large language models): 15 → 67 papers
- Gaussian splatting (3D vision): 2 → 45 papers
- Language, safety and interpretability (Large language models): 121 → 240 papers
- Video generation (Video): 25 → 82 papers
- Video-language understanding (Video): 6 → 48 papers
- Attention and transformers (Architectures and efficiency): 66 → 140 papers
More venues
Same year: ICML 2025 · NeurIPS 2025 · CVPR 2025 · AAAI 2025