Compression AI Lab
See AI through the lens of Compression
AI is Compression
We view intelligence as the ability to compress vast amounts of information into compact representations and, from that compression, recover information and derive new insights.
Thinking via Compression
We apply the same principle to how we think. We compress complex problems by questioning assumptions and keeping only what must be true, then reason from that core to derive new insights.
Representation
Represent more with less
Efficient Attention Mechanisms · Model & Token Pruning · Learned Image Compression & Super-Resolution
Learning
Learn what matters
Long-Tailed Learning · Vision-Language Alignment · Human-Aligned Perception
Robustness
Remain effective under change
Distribution Shift · Adversarial Robustness · Reliable Evaluation
Interpretability
Understand what drives model decisions
Visual Attribution · ViT Explainability · VLM Explainability
News
| Aug. 2026 | 1 paper accepted to ECCV 2026 Workshop! 🎉 |
|---|---|
| Apr. 2026 | 1 paper accepted to ICIP 2026! 🎉 |
| Mar. 2026 | CAI Lab received the NRF Outstanding Young Scientist Grant. 🎉 |
| Apr. 2025 | CAI Lab’s homepage unveiled!  |
| Mar. 2025 | Jun-Hyuk Kim joined the Department of AI at Chung-Ang University as an Assistant Professor.  |