# No Training Wheels: Steering Vectors for Bias Correction at Inference Time

arXiv: https://arxiv.org/abs/2506.18598 · HTML: https://mlsanity.com/paper/2506.18598/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Probing deception, bias and beliefs, one of 13 directions in Mechanistic interpretability.
- Builds on 14 of its 17 references that are in the map.
- First posted on arXiv 23 Jun 2025.

## Placed in

- Mechanistic interpretability / [Probing deception, bias and beliefs](https://mlsanity.com/mechinterp/probing-deception-bias-and-beliefs/)

## Builds on (most cited first, 14 of 14)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [A Broad-Coverage Challenge Corpus for Sentence Understanding through Inference](https://mlsanity.com/paper/1704.05426/) (2017)
- [Shortcut Learning in Deep Neural Networks](https://mlsanity.com/paper/2004.07780/) (2020)
- [Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings](https://mlsanity.com/paper/1607.06520/) (2016)
- [Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization](https://mlsanity.com/paper/1911.08731/) (2019)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [Discovering Latent Knowledge in Language Models Without Supervision](https://mlsanity.com/paper/2212.03827/) (2022)
- [The Linear Representation Hypothesis and the Geometry of Large Language Models](https://mlsanity.com/paper/2311.03658/) (2023)
- [Annotation Artifacts in Natural Language Inference Data](https://mlsanity.com/paper/1803.02324/) (2018)
- [Age Progression/Regression by Conditional Adversarial Autoencoder](https://mlsanity.com/paper/1702.08423/) (2017)
- [Linear Representations of Sentiment in Large Language Models](https://mlsanity.com/paper/2310.15154/) (2023)
- [CelebA-Spoof: Large-Scale Face Anti-Spoofing Dataset with Rich Annotations](https://mlsanity.com/paper/2007.12342/) (2020)

## Built on by (most cited first, 2 of 2)

- [Fairness Evaluation and Inference Level Mitigation in LLMs](https://mlsanity.com/paper/2510.18914/) (2025)
- [Backdoor Directions in Vision Transformers](https://mlsanity.com/paper/2603.10806/) (2026)

## Abstract (arXiv)

Neural network classifiers trained on datasets with uneven group representation often inherit class biases and learn spurious correlations. These models may perform well on average but consistently fail on atypical groups. For example, in hair color classification, datasets may over-represent females with blond hair, reinforcing stereotypes. Although various algorithmic and data-centric methods have been proposed to address such biases, they often require retraining or significant compute. In this work, we propose a cheap, training-free method inspired by steering vectors used to edit behaviors in large language models. We compute the difference in mean activations between majority and minority groups to define a "bias vector," which we subtract from the model's residual stream. This leads to reduced classification bias and improved worst-group accuracy. We explore multiple strategies for extracting and applying these vectors in transformer-like classifiers, showing that steering vectors, traditionally used in generative models, can also be effective in classification. More broadly, we showcase an extremely cheap, inference time, training free method to mitigate bias in classification models.