ICML 2023 accepted papers by research area

1,865 papers were accepted at ICML 2023 (+632 vs 2022). Here is how they split across 20 research areas, and which topics gained the most ground.

Research areas

Area	Papers	Share	Change in share vs 2022
Learning theory and algorithms	283	15.2%	-0.4 pts
Probabilistic and scientific modeling	254	13.6%	+0.7 pts
Reinforcement learning	191	10.2%	+0.1 pts
Label-efficient and robust learning	181	9.7%	-1.4 pts
Architectures and efficiency	143	7.7%	-1.7 pts
Optimization	135	7.2%	-1.4 pts
Graph learning	116	6.2%	+0.9 pts
Bandits and online learning	105	5.6%	-0.5 pts
Federated learning and privacy	90	4.8%	-0.1 pts
Image generation	76	4.1%	+2.5 pts
Adversarial robustness and security	56	3.0%	-1.6 pts
Large language models	44	2.4%	+1.1 pts
Multi-agent systems and games	44	2.4%	steady
Vision-language and multimodal	41	2.2%	+0.8 pts
Time series	28	1.5%	+0.4 pts
Detection and segmentation	25	1.3%	+0.3 pts
Continual learning	21	1.1%	steady
3D vision	12	0.6%	+0.2 pts
Video	11	0.6%	+0.3 pts
Point clouds	9	0.5%	steady

Change in share of ICML papers from 2022 to 2023, topics with at least 10 papers.

Fast diffusion sampling (Image generation): 4 → 34 papers
GNN architectures (Graph learning): 28 → 60 papers
Variational and generative sampling (Probabilistic and scientific modeling): 37 → 70 papers
Visual and representation RL (Reinforcement learning): 33 → 63 papers
Treatment effects (Probabilistic and scientific modeling): 8 → 24 papers
Online convex optimization and games (Bandits and online learning): 24 → 48 papers
Autoregressive and latent generation (Image generation): 7 → 21 papers
Sample complexity (Learning theory and algorithms): 27 → 51 papers
Prediction and decision losses (Learning theory and algorithms): 49 → 83 papers
Combinatorial and search optimization (Learning theory and algorithms): 10 → 24 papers

ICML in other years: 2024 · 2025 · 2026