# Attribution Graphs and Causal Probing for Mechanistic Discovery and Bias Repair in Multimodal Generative Learning

Preprint · Noor Islam S. Mohammad, Uluğ Bayazıt

- Archive id: arxiv:2510.12957
- Published: 2025-10-14
- arXiv: https://arxiv.org/abs/2510.12957
- Citations: 0
- Source URL: https://arxiv.org/abs/2510.12957
- PDF: https://arxiv.org/pdf/2510.12957
- Provenance: arxiv

## Abstract

We treat the internals of generative models as mechanistic objects rather than black boxes. We introduce \\textbf{Attribution Graphs} (AGs), which extend GradCAM++ to circuit-level representations, and \\textbf{Causal Probing}, a do-calculus intervention method for identifying causal latent structures, enabling detection and correction of spurious correlations, demographic biases, and misaligned decision circuits during training. We further propose the \\textbf{Cognitive Alignment Score (CAS)}, quantifying agreement between model-internal representations and human concepts, a \\textbf{saliency-first privacy mechanism} sharing only thresholded attribution nodes, a bias-aware regularizer aligning subgroup statistics, and a Reveal-to-Revise loop integrating attribution signals into parameter updates without separate fine-tuning. Evaluated on CelebA, FairFace, Jigsaw, and HateXplain, our method achieves \\textbf{94.1\\%} accuracy, \\textbf{92.3\\%} macro F1, \\textbf{79.4\\%} IoU-XAI, and \\textbf{12.7} FID at 72--76\\% adversarial robustness, while reducing subgroup disparity $Δ\_{\\mathrm{bias}}$ by \\textbf{41\\%}, demonstrating that mechanistic interpretability, fairness, and generative performance can be jointly optimized.

_Abstract reproduced under CC0-1.0 (arXiv descriptive metadata)._

Topics: [Mechanistic Interpretability](https://space-kitty.com/topics/interpretability)

---
Space Kitty Archive · https://space-kitty.com · API: https://space-kitty.com/api/v1 · MCP: https://space-kitty.com/mcp · Premium: $0.20/request via x402 (Base Sepolia (testnet)). Metadata: arXiv (CC0), OpenAlex (CC0), Europe PMC (CC BY/CC0 works), ClinicalTrials.gov.