NeurIPS 2025 Explorer

5896 paper abstracts · 4246 concepts · 2217 prolific authors · 1645-term glossary. Built to read a conference the way you read a knowledge graph: start anywhere, follow the links.

Most-mentioned concepts

• reinforcement learning (384)
• generalization (242)
• computational efficiency (202)
• fine-tuning (198)
• diffusion models (195)
• vision-language models (173)
• robustness (170)
• state-of-the-art performance (160)
• theoretical analysis (140)
• multimodal large language models (128)
• generative models (119)
• interpretability (112)
• supervised fine-tuning (101)
• state-of-the-art methods (100)
• scalability (100)
• performance improvement (95)
• empirical results (95)
• benchmark datasets (91)
• performance evaluation (89)
• foundation models (86)
• computational overhead (85)
• graph neural networks (80)
• overfitting (79)
• model performance (78)
• performance degradation (73)
• real-world datasets (71)
• sample complexity (70)
• transformers (70)
• self-supervised learning (67)
• experimental results (66)
• in-context learning (66)
• benchmarks (64)
• downstream tasks (64)
• synthetic data (63)
• empirical validation (63)
• computational cost (62)
• large language model (62)
• sample efficiency (62)
• representation learning (61)
• benchmark evaluation (61)
• theoretical guarantees (60)
• empirical evaluation (60)
• empirical evaluations (59)
• uncertainty quantification (57)
• generative modeling (57)
• computational complexity (57)
• numerical experiments (56)
• synthetic datasets (55)
• convergence (55)
• evaluation metrics (54)
• federated learning (54)
• contrastive learning (53)
• optimization (52)
• gradient descent (51)
• training dynamics (50)
• deep neural networks (50)
• latent space (50)
• ablation studies (49)
• reasoning capabilities (49)
• computational costs (48)