Does Mechanistic Interpretability Transfer Across Data Modalities? A Cross-Domain Causal Circuit Analysis of Variational Autoencoders

• URL: http://arxiv.org/abs/2603.21236v1
• Date: Sun, 22 Mar 2026 13:51:45 GMT
• Title: Does Mechanistic Interpretability Transfer Across Data Modalities? A Cross-Domain Causal Circuit Analysis of Variational Autoencoders
• Abstract summary: Tabular image-related variational autoencoders (VAEs) are increasingly used for imputation, anomaly detection, and synthetic data generation.<n>This paper introduces three new techniques: posterior-calibration of Causal Effect Strength (CES), path-specific activation patching, and Feature-Group Disentanglement (FGD)
• Score: 6.5607014897534865
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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. In addition to extending a four-level causal intervention framework to four tabular and one image benchmark across five different VAE architectures (with 75 individual training runs per architecture and three random seed values for each run), this paper introduces three new techniques: posterior-calibration of Causal Effect Strength (CES), path-specific activation patching, and Feature-Group Disentanglement (FGD). The results from our experiments demonstrate that: (i) Tabular VAEs have circuits with modularity that is approximately 50% lower than their image counterparts. (ii) $β$-VAE experiences nearly complete collapse in CES scores when applied to heterogeneous tabular features (0.043 CES score for tabular data compared to 0.133 CES score for images), which can be directly attributed to reconstruction quality degradation (r = -0.886 correlation coefficient between CES and MSE). (iii) CES successfully captures nine of eleven statistically significant architecture differences using Holm--Šidák corrections. (iv) Interventions with high specificity predict the highest downstream AUC values (r = 0.460, p < .001). This study challenges the common assumption that architectural guidance from image-related studies can be transferred to tabular datasets.

Related papers

• Architecture-Dependent Causal Transfer of Activation States Across Large Language Models [51.56484100374058]
  Internal language activation states can be transferred causally between different large language models.<n>Injecting projected activations into a target model during generation produces a statistically significant, preregistered causal effect.<n>We conclude that end-to-end activation-state transfer between LLMs, as currently implemented, is architecture-dependent rather than universal.
  arXiv  Detail & Related papers  (2026-08-17T09:53:27Z)
• CIPHER: Causal Intervention Pathways for Healthcare Equity and Robustness [12.8473199678431]
  We formalize the medical image formation process via a structural causal model.<n>We introduce CIPHER, a framework designed to intervene on all four causal paths.<n>By employing a multi-pathway intervention strategy, our model reduced worst-group disparities by an average of 35.8%.
  arXiv  Detail & Related papers  (2026-07-01T09:36:53Z)
• HypCBC: Domain-Invariant Hyperbolic Cross-Branch Consistency for Generalizable Medical Image Analysis [1.8747639074211104]
  We present the first comprehensive validation of hyperbolic representation learning for medical image analysis.<n>We demonstrate statistically significant gains across eleven in-distribution datasets and three ViT models.<n>Our proposed method promotes domain-invariant features and outperforms state-of-the-art Euclidean methods by an average of $+2.1%$ AUC.
  arXiv  Detail & Related papers  (2026-02-03T08:50:24Z)
• ShortcutBreaker: Low-Rank Noisy Bottleneck with Global Perturbation Attention for Multi-Class Unsupervised Anomaly Detection [59.89803740308262]
  ShortcutBreaker is a novel unified feature-reconstruction framework for MUAD tasks.<n>It features two key innovations to address the issue of shortcuts.<n>The proposed method achieves a remarkable image-level AUROC of 99.8%, 98.9%, 90.6%, and 87.8% on four datasets.
  arXiv  Detail & Related papers  (2025-10-21T06:51:30Z)
• Adapting HFMCA to Graph Data: Self-Supervised Learning for Generalizable fMRI Representations [57.054499278843856]
  Functional magnetic resonance imaging (fMRI) analysis faces significant challenges due to limited dataset sizes and domain variability between studies.<n>Traditional self-supervised learning methods inspired by computer vision often rely on positive and negative sample pairs.<n>We propose adapting a recently developed Hierarchical Functional Maximal Correlation Algorithm (HFMCA) to graph-structured fMRI data.
  arXiv  Detail & Related papers  (2025-10-05T12:35:01Z)
• MoCap-Impute: A Comprehensive Benchmark and Comparative Analysis of Imputation Methods for IMU-based Motion Capture Data [4.498558555177452]
  Motion capture (MoCap) data from wearable Inertial Measurement Units (IMUs) is vital for applications in sports science.<n>Despite numerous imputation techniques, a systematic performance evaluation for IMU-derived MoCap time-series data is lacking.
  arXiv  Detail & Related papers  (2025-07-14T14:41:19Z)
• Pose Matters: Evaluating Vision Transformers and CNNs for Human Action Recognition on Small COCO Subsets [0.0]
  This study explores human recognition using a three-class subset of the COCO image corpus.<n>The binary Vision Transformer (ViT) achieved 90% mean test accuracy.
  arXiv  Detail & Related papers  (2025-06-13T11:16:50Z)
• Causal Intervention Framework for Variational Auto Encoder Mechanistic Interpretability [0.0]
  This paper introduces a comprehensive causal intervention framework for mechanistic interpretability of Variational Autoencoders (VAEs)<n>We develop techniques to identify and analyze "circuit motifs" in VAEs, examining how semantic factors are encoded, processed, and disentangled through the network layers.<n>Results show that our interventions can successfully isolate functional circuits, map computational graphs to causal graphs of semantic factors, and distinguish between polysemantic and monosemantic units.
  arXiv  Detail & Related papers  (2025-05-06T13:40:59Z)
• OCT-SelfNet: A Self-Supervised Framework with Multi-Modal Datasets for Generalized and Robust Retinal Disease Detection [2.3349787245442966]
  Our research contributes a self-supervised robust machine learning framework, OCT-SelfNet, for detecting eye diseases. Our method addresses the issue using a two-phase training approach that combines self-supervised pretraining and supervised fine-tuning. In terms of the AUC-PR metric, our proposed method exceeded 42%, showcasing a substantial increase of at least 10% in performance compared to the baseline.
  arXiv  Detail & Related papers  (2024-01-22T20:17:14Z)
• The effect of data augmentation and 3D-CNN depth on Alzheimer's Disease detection [51.697248252191265]
  This work summarizes and strictly observes best practices regarding data handling, experimental design, and model evaluation. We focus on Alzheimer's Disease (AD) detection, which serves as a paradigmatic example of challenging problem in healthcare. Within this framework, we train predictive 15 models, considering three different data augmentation strategies and five distinct 3D CNN architectures.
  arXiv  Detail & Related papers  (2023-09-13T10:40:41Z)
• Source-Free Collaborative Domain Adaptation via Multi-Perspective Feature Enrichment for Functional MRI Analysis [55.03872260158717]
  Resting-state MRI functional (rs-fMRI) is increasingly employed in multi-site research to aid neurological disorder analysis. Many methods have been proposed to reduce fMRI heterogeneity between source and target domains. But acquiring source data is challenging due to concerns and/or data storage burdens in multi-site studies. We design a source-free collaborative domain adaptation framework for fMRI analysis, where only a pretrained source model and unlabeled target data are accessible.
  arXiv  Detail & Related papers  (2023-08-24T01:30:18Z)
• Vision Transformers are Robust Learners [65.91359312429147]
  We study the robustness of the Vision Transformer (ViT) against common corruptions and perturbations, distribution shifts, and natural adversarial examples. We present analyses that provide both quantitative and qualitative indications to explain why ViTs are indeed more robust learners.
  arXiv  Detail & Related papers  (2021-05-17T02:39:22Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.