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

2025/10/14 by Mohammad, Noor Islam S., Bayazıt, Uluğ
Computer Science · #Adversarial Robustness in Machine Learning #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · pdf · doi:10.48550/arxiv.2510.12957

openalex publication_date 2025/10/14 · openalex created_date 2025/10/17 · openalex updated_date 2026/07/31

Abstract

We treat the internals of generative models as mechanistic objects rather than black boxes. We introduce Attribution Graphs (AGs), which extend GradCAM++ to circuit-level representations, and 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 Cognitive Alignment Score (CAS), quantifying agreement between model-internal representations and human concepts, a 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 94.1% accuracy, 92.3% macro F1, 79.4% IoU-XAI, and 12.7 FID at 72--76% adversarial robustness, while reducing subgroup disparity Δ_bias by 41%, demonstrating that mechanistic interpretability, fairness, and generative performance can be jointly optimized.

Citations

• Reveal to Revise: An Explainable AI Life Cycle for Iterative Bias Correction of Deep Models
• Grounded Copilot: How Programmers Interact with Code-Generating Models
• Hierarchical Text-Conditional Image Generation with CLIP Latents
• On the Opportunities and Risks of Foundation Models
• On the Dangers of Stochastic Parrots
• HateXplain: A Benchmark Dataset for Explainable Hate Speech Detection
• Language Models are Few-Shot Learners
• Towards Faithfully Interpretable NLP Systems: How should we define and\n evaluate faithfulness?
• Fooling LIME and SHAP: Adversarial Attacks on Post hoc Explanation\n Methods
• Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI
• A Survey on Bias and Fairness in Machine Learning
• FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age
• DARPA's Explainable Artificial Intelligence Program
• Explanations can be manipulated and geometry is to blame
• Multimodal Transformer for Unaligned Multimodal Language Sequences
• Ablation Studies in Artificial Neural Networks
• Robustness May Be at Odds with Accuracy
• Generative Adversarial Networks: An Overview
• Deep Reinforcement Learning that Matters
• Towards Deep Learning Models Resistant to Adversarial Attacks
• Attention Is All You Need
• A Unified Approach to Interpreting Model Predictions
• Network Dissection: Quantifying Interpretability of Deep Visual\n Representations
• Towards A Rigorous Science of Interpretable Machine Learning
• Image-to-Image Translation with Conditional Adversarial Networks
• Adversarial Machine Learning at Scale
• Conditional Image Synthesis With Auxiliary Classifier GANs
• Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization
• Improved Techniques for Training GANs
• "Why Should I Trust You?": Explaining the Predictions of Any Classifier
• Deep Residual Learning for Image Recognition
• Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning
• Explaining and Harnessing Adversarial Examples
• Deep Learning Face Attributes in the Wild
• Conditional Generative Adversarial Nets
• Neural Machine Translation by Jointly Learning to Align and Translate
• Practical Bayesian Optimization of Machine Learning Algorithms
• BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

Related

• Focus! Rating XAI Methods and Finding Biases
• A Review of Multimodal Explainable Artificial Intelligence: Past, Present and Future
• Reliable or Deceptive? Investigating Gated Features for Smooth Visual Explanations in CNNs
• How can we trust opaque systems? Criteria for robust explanations in XAI
• Precise Benchmarking of Explainable AI Attribution Methods