Fairness is Not Flat: Geometric Phase Transitions Against Shortcut Learning

TL;DR

This paper introduces a geometric methodology using a Topological Auditor to mitigate shortcut learning in neural networks, promoting ethical representations and reducing demographic biases efficiently.

Contribution

It presents a novel geometric phase transition approach that isolates shortcut features and encourages higher capacity learning to improve fairness and robustness.

Findings

Pruning linear shortcuts forces networks to use higher geometric capacity.

The method outperforms L1 regularization in reducing demographic bias.

Reduces counterfactual gender vulnerability from 21.18% to 7.66%.

Abstract

Deep Neural Networks are highly susceptible to shortcut learning, frequently memorizing low-dimensional spurious correlations instead of underlying causal mechanisms. This phenomenon not only degrades out-of-distribution robustness but also induces severe demographic biases in sensitive applications. In this paper, we propose a geometric \textit{a priori} methodology to mitigate shortcut learning. By deploying a zero-hidden-layer ($N=1$) Topological Auditor, we mathematically isolate features that monopolize the gradient without human intervention. We empirically demonstrate a Capacity Phase Transition: once linear shortcuts are pruned, networks are forced to utilize higher geometric capacity ($N \geq 16$) to curve the decision boundary and learn ethical representations. Our approach outperforms L1 Regularization -- which collapses into demographic bias -- and operates at a fraction of the computational cost of post-hoc methods like Just Train Twice (JTT), successfully reducing counterfactual gender vulnerability from 21.18\% to 7.66\%.