Counterfactual Explanations on Robust Perceptual Geodesics

2026-01-26

AI Summary

Existing counterfactual explanation methods often employ inappropriate distance metrics—such as Euclidean distance—leading to semantic drift, off-manifold artifacts, or adversarial collapse, which hinder the generation of human-perceptually plausible explanations. To address this, this work proposes Perceptual Counterfactual Geodesics (PCG), a method that constructs a perceptually aligned Riemannian metric using robust visual features and optimizes counterfactual samples along geodesics in the latent space under this metric. This ensures that generated explanations are semantically consistent, lie on the data manifold, and vary smoothly. Experiments across three visual datasets demonstrate that PCG significantly outperforms existing baselines, producing more credible explanations and uncovering model failure modes obscured under standard distance metrics.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Humans and AI: Explainable AI (XAI) for Human Understanding
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Latent-space optimization methods for counterfactual explanations - framed as minimal semantic perturbations that change model predictions - inherit the ambiguity of Wachter et al.'s objective: the choice of distance metric dictates whether perturbations are meaningful or adversarial. Existing approaches adopt flat or misaligned geometries, leading to off-manifold artifacts, semantic drift, or adversarial collapse. We introduce Perceptual Counterfactual Geodesics (PCG), a method that constructs counterfactuals by tracing geodesics under a perceptually Riemannian metric induced from robust vision features. This geometry aligns with human perception and penalizes brittle directions, enabling smooth, on-manifold, semantically valid transitions. Experiments on three vision datasets show that PCG outperforms baselines and reveals failure modes hidden under standard metrics.

Problem

• counterfactual explanations
• latent-space optimization
• perceptual geometry
• semantic perturbations
• on-manifold generation

Innovation

• counterfactual explanations
• perceptual geodesics
• Riemannian metric
• on-manifold generation
• robust vision features