[Paper] Counterfactual Explanations on Robust Perceptual Geodesics

Source: arXiv - 2601.18678v1

Overview

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.

Key Contributions

• cs.LG
• cs.CV
• cs.HC
• math.DG

Methodology

Please refer to the full paper for detailed methodology.

Practical Implications

This research contributes to the advancement of cs.LG.

Authors

• Eslam Zaher
• Maciej Trzaskowski
• Quan Nguyen
• Fred Roosta

Paper Information

• arXiv ID: 2601.18678v1
• Categories: cs.LG, cs.CV, cs.HC, math.DG
• Published: January 26, 2026
• PDF: Download PDF