AI RESEARCH
Does Mechanistic Interpretability Transfer Across Data Modalities? A Cross-Domain Causal Circuit Analysis of Variational Autoencoders
arXiv CS.LG
•
ArXi:2603.21236v1 Announce Type: new Although mechanism-based interpretability has generated an abundance of insight for discriminative network analysis, generative models are less understood -- particularly outside of image-related applications. We investigate how much of the causal circuitry found within image-related variational autoencoders (VAEs) will generalize to tabular data, as VAEs are increasingly used for imputation, anomaly detection, and synthetic data generation.