New neural architecture CauScale enables efficient causal discovery at scale

Researchers have developed CauScale, a novel neural architecture designed to overcome the efficiency limitations of existing causal discovery methods when applied to large-scale graphs. CauScale employs a data compression unit and tied attention weights to significantly improve time and space efficiency, enabling inference on graphs with up to 1000 nodes. The architecture utilizes a two-stream design, combining a data stream for relational evidence extraction and a graph stream for integrating statistical priors and structural signals. This approach allows CauScale to achieve high accuracy, with 99.6% mAP on in-distribution data and 84.4% on out-of-distribution data, while offering substantial inference speedups compared to previous methods. AI

IMPACT This research could accelerate scientific AI and data analysis by enabling more efficient causal discovery on larger datasets.

RANK_REASON The cluster contains an academic paper detailing a new method for neural causal discovery. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• arXiv
• Bo Peng
• CauScale

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →