A Foundation-style Model for Zero-Shot Statistical Dependency Measurement
Published in International Conference on Machine Learning (ICML), 2026
Recommended citation: Zhengyang Hu*, Yanzhi Chen*, Hanxiang Ren, Qunsong Zeng, Youyi Zheng, Adrian Weller, Kaibin Huang, Yanchao Yang. (2026). "A Foundation-style Model for Zero-Shot Statistical Dependency Measurement." ICML 2026. (*Equal contribution.)
InfoAtlas is a full upgrade of InfoNet that extends the foundation-model paradigm to arbitrary-dimensional statistical dependency measurement. We redesign both the architecture and the synthetic pre-training distribution, and scale the model to ~1B parameters by pre-training for approximately one month on a 32xH200 GPU cluster. The resulting model is zero-shot ready and demonstrates strong performance across a substantially broader set of downstream tasks than InfoNet.
Authors: Zhengyang Hu*, Yanzhi Chen*, Hanxiang Ren, Qunsong Zeng, Youyi Zheng, Adrian Weller, Kaibin Huang, Yanchao Yang. (*Equal contribution.)