GRADIEND: Feature Learning within Neural Networks Exemplified through Biases

Jonathan Drechsel, Steffen Herbold
2/3/2025
cs.LGcs.AIcs.CL

No official code found. Best implementation: aieng-lab/gradiend-bias (3 stars, Apache-2.0, last commit 2026-03-03). 1 more below.

Abstract

AI systems frequently exhibit and amplify social biases, leading to harmful consequences in critical areas. This study introduces a novel encoder-decoder approach that leverages model gradients to learn a feature neuron encoding societal bias information such as gender, race, and religion. We show that our method can not only identify which weights of a model need to be changed to modify a feature, but even demonstrate that this can be used to rewrite models to debias them while maintaining other capabilities. We demonstrate the effectiveness of our approach across various model architectures and highlight its potential for broader applications.

View on arXivView PDF

Code Implementations(2)

aieng-lab/gradiendCommunity100%
Apache-2.0

GRADIEND Python Package

21Shell, PythonFeb 10, 20267 months agoApache-2.0
aieng-lab/gradiend-biasCommunity67%
Apache-2.0

GRADIEND: Feature Learning within neural networks exemplified through biases (ICLR 2026)

32Shell, PythonJan 30, 20257 months agoApache-2.0

Similar Papers

Stereotype Detection as a Catalyst for Enhanced Bias Detection: A Multi-Task Learning Approach

85% match2025

Shielded Representations: Protecting Sensitive Attributes Through Iterative Gradient-Based Projection

84% match2023

The Devil is in the Neurons: Interpreting and Mitigating Social Biases in Pre-trained Language Models

83% match2024

Interpretable Debiasing of Vision-Language Models for Social Fairness

83% match2026

Biased Attention: Do Vision Transformers Amplify Gender Bias More than Convolutional Neural Networks?

83% match2023

From Prejudice to Parity: A New Approach to Debiasing Large Language Model Word Embeddings

82% match2024

Cite this paper

@article{drechsel2025gradiend,
  title  = {GRADIEND: Feature Learning within Neural Networks Exemplified through Biases},
  author = {Jonathan Drechsel and Steffen Herbold},
  year   = {2025},
  eprint = {2502.01406},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2502.01406v3}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2502.01406.svg)](https://codeofpaper.com/paper/2502.01406)

Discussion