# Mechanistic Interpretability of Fine-Tuned Vision Transformers on Distorted Images: Decoding Attention Head Behavior for Transparent and Trustworthy AI

arXiv: https://arxiv.org/abs/2503.18762 · HTML: https://mlsanity.com/paper/2503.18762/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in Circuits and mechanistic analysis, one of 13 directions in Mechanistic interpretability.
- Builds on 6 of its 9 references that are in the map.
- First posted on arXiv 24 Mar 2025.

## Placed in

- Mechanistic interpretability / [Circuits and mechanistic analysis](https://mlsanity.com/mechinterp/circuits-and-mechanistic-analysis/)

## Builds on (most cited first, 6 of 6)

- [How to use and interpret activation patching](https://mlsanity.com/paper/2404.15255/) (2024)
- [From Explainable to Interpretable Deep Learning for Natural Language Processing in Healthcare: How Far from Reality?](https://mlsanity.com/paper/2403.11894/) (2024)
- [In-Context Linear Regression Demystified: Training Dynamics and Mechanistic Interpretability of Multi-Head Softmax Attention](https://mlsanity.com/paper/2503.12734/) (2025)
- [Interpretable Neural Causal Models with TRAM-DAGs](https://mlsanity.com/paper/2503.16206/) (2025)
- [Towards Explainable Privacy Preservation in Federated Learning via Shapley Value-Guided Noise Injection](https://mlsanity.com/paper/2503.12958/) (2025)
- [A Dual-Directional Context-Aware Test-Time Learning for Text Classification](https://mlsanity.com/paper/2503.15469/) (2025)

## Built on by (most cited first, 5 of 5)

- [Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models](https://mlsanity.com/paper/2601.14004/) (2026)
- [Understanding Geometric Representations in Self-Supervised Vision Transformers via Subspace Intervention](https://mlsanity.com/paper/2607.01987/) (2026)
- [Vision Transformers Exhibit Human-Like Biases: Evidence of Orientation and Color Selectivity, Categorical Perception, and Phase Transitions](https://mlsanity.com/paper/2504.09393/) (2025)
- [DARTS-GT: Differentiable Architecture Search for Graph Transformers with Quantifiable Instance-Specific Interpretability Analysis](https://mlsanity.com/paper/2510.14336/) (2025)
- [Sparse but not Simpler: A Multi-Level Interpretability Analysis of Vision Transformers](https://mlsanity.com/paper/2603.15919/) (2026)

## Abstract (arXiv)

Mechanistic interpretability improves the safety, reliability, and robustness of large AI models. This study examined individual attention heads in vision transformers (ViTs) fine tuned on distorted 2D spectrogram images containing non relevant content (axis labels, titles, color bars). By introducing extraneous features, the study analyzed how transformer components processed unrelated information, using mechanistic interpretability to debug issues and reveal insights into transformer architectures. Attention maps assessed head contributions across layers. Heads in early layers (1 to 3) showed minimal task impact with ablation increased MSE loss slightly (μ=0.11%, σ=0.09%), indicating focus on less critical low level features. In contrast, deeper heads (e.g., layer 6) caused a threefold higher loss increase (μ=0.34%, σ=0.02%), demonstrating greater task importance. Intermediate layers (6 to 11) exhibited monosemantic behavior, attending exclusively to chirp regions. Some early heads (1 to 4) were monosemantic but non task relevant (e.g. text detectors, edge or corner detectors). Attention maps distinguished monosemantic heads (precise chirp localization) from polysemantic heads (multiple irrelevant regions). These findings revealed functional specialization in ViTs, showing how heads processed relevant vs. extraneous information. By decomposing transformers into interpretable components, this work enhanced model understanding, identified vulnerabilities, and advanced safer, more transparent AI.