Sparse Autoencoders enable Robust and Interpretable Fine-tuning of CLIP models
For practitioners fine-tuning vision-language models, SAE-FT offers a computationally efficient and interpretable method to maintain robustness against distribution shifts.
SAE-FT uses sparse autoencoders to regularize fine-tuning of CLIP's visual representations, preventing catastrophic forgetting and improving robustness. It matches or exceeds state-of-the-art performance on ImageNet and distribution shift benchmarks.
Large-scale pre-trained vision-language models like CLIP demonstrate remarkable zero-shot performance across diverse tasks. However, fine-tuning these models to improve downstream performance often degrades robustness against distribution shifts. Recent approaches have attempted to mitigate this trade-off, but often rely on computationally expensive text-guidance. We propose a novel method for robust fine-tuning, SAE-FT, which operates only on the model's visual representations. SAE-FT regularizes changes to these representations by penalizing the addition and removal of semantically meaningful features identified by a Sparse Autoencoder trained on the pre-trained model. This constraint prevents catastrophic forgetting and makes the fine-tuning process interpretable, enabling direct analysis of semantic changes. SAE-FT is both mechanistically transparent and computationally efficient, matching or exceeding state-of-the-art performance on ImageNet and its associated distribution shift benchmarks. Code is publicly available at: https://github.com/Fabian-Mor/sae-ft.