Performance degradation on Nonlinear dataset with higher dimensions (d=20/40)

[zhangzzzzz1]

Hi, thank you for sharing this interesting work!

I have a question regarding the Nonlinear dataset experiments. In the paper, the Nonlinear dataset is evaluated with 6 dimensions. I'm wondering whether you have tested AERCA on higher-dimensional settings of the Nonlinear dataset (e.g., 20 or 40 dimensions)?

I tried running AERCA on the Nonlinear dataset with higher dimensions (20 and 40), but the performance dropped significantly compared to the 6-dimensional results reported in the paper. Specifically, both the causal discovery metrics and root cause identification accuracy degraded noticeably.

Could you shed some light on:

1. Whether high-dimensional Nonlinear settings were tested internally?
2. Any recommended hyperparameter tuning or architectural adjustments for scaling to higher dimensions?

Thank you!