# Steering LVLMs via Sparse Autoencoder for Hallucination Mitigation

arXiv: https://arxiv.org/abs/2505.16146 · HTML: https://mlsanity.com/paper/2505.16146/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Placed in Hallucination mitigation in LVLMs, one of 16 directions in Multimodal LLMs / vision-language models.
- First posted on arXiv 22 May 2025.

## Placed in

- Multimodal LLMs / vision-language models / [Hallucination mitigation in LVLMs](https://mlsanity.com/mllm/hallucination-mitigation-in-lvlms/)
- Mechanistic interpretability / [Hallucination detection from model internals](https://mlsanity.com/mechinterp/hallucination-detection-from-model-internals/)

## Built on by (most cited first, 6 of 6)

- [HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-Encoders](https://mlsanity.com/paper/2604.16430/) (2026)
- [DO-Bench: An Attributable Benchmark for Diagnosing Object Hallucination in Vision-Language Models](https://mlsanity.com/paper/2604.22822/) (2026)
- [Knowledge Vector of Logical Reasoning in Large Language Models](https://mlsanity.com/paper/2604.23877/) (2026)
- [Rethinking Visual Neglect: Steering via Context-Preference for MLLM Hallucination Mitigation](https://mlsanity.com/paper/2605.27993/) (2026)
- [Decodable but Misrouted: Sparse Features Uncover a Readout Gap in Vision-Language Models for Harmful Meme Detection](https://mlsanity.com/paper/2609.18860/) (2026)
- [NeuronEye: Query-Guided Visual Concept Activation for Vision-Language Reasoning](https://mlsanity.com/paper/2609.38098/) (2026)

## Abstract (arXiv)

Large vision-language models (LVLMs) have achieved remarkable performance on multimodal tasks. However, they still suffer from hallucinations, generating text inconsistent with visual input, posing significant risks in real-world applications. Existing approaches to address this issue focus on incorporating external knowledge bases, alignment training, or decoding strategies, all of which require substantial computational cost and time. Recent works try to explore more efficient alternatives by adjusting LVLMs' internal representations. Although promising, these methods may cause hallucinations to be insufficiently suppressed or lead to excessive interventions that negatively affect normal semantics. In this work, we leverage sparse autoencoders (SAEs) to identify semantic directions closely associated with faithfulness or hallucination, extracting more precise and disentangled hallucination-related representations. Our analysis demonstrates that interventions along the identified faithful direction can mitigate hallucinations, while those along the hallucinatory direction can exacerbate them. Building on these insights, we propose Steering LVLMs via SAE Latent Directions (SSL), a plug-and-play method based on SAE-derived latent directions to mitigate hallucinations in LVLMs. Extensive experiments demonstrate that SSL significantly outperforms existing decoding approaches in mitigating hallucinations, while maintaining transferability across different model architectures with negligible additional time overhead. The code is available at https://github.com/huazhenglin2003/SSL.