IMSE: Intrinsic Mixture of Spectral Experts Fine-tuning for Test-Time Adaptation
This work addresses the problem of efficiently adapting large pretrained models to new test distributions for practitioners and researchers working with distribution shifts, offering significant gains with minimal parameter updates.
This paper introduces Intrinsic Mixture of Spectral Experts (IMSE) for test-time adaptation (TTA) in Vision Transformers, which adapts only singular values from SVD-decomposed linear layers. IMSE achieves state-of-the-art performance in TTA and improves accuracy by 3.4 percentage points (pp) in continual TTA (CTTA) and 2.4 pp in gradual CTTA, while using 385 times fewer trainable parameters.
Test-time adaptation (TTA) has been widely explored to prevent performance degradation when test data differ from the training distribution. However, fully leveraging the rich representations of large pretrained models with minimal parameter updates remains underexplored. In this paper, we propose Intrinsic Mixture of Spectral Experts (IMSE) that leverages the spectral experts inherently embedded in Vision Transformers. We decompose each linear layer via singular value decomposition (SVD) and adapt only the singular values, while keeping the singular vectors fixed. We further identify a key limitation of entropy minimization in TTA: it often induces feature collapse, causing the model to rely on domain-specific features rather than class-discriminative features. To address this, we propose a diversity maximization loss based on expert-input alignment, which encourages diverse utilization of spectral experts during adaptation. In the continual test-time adaptation (CTTA) scenario, beyond preserving pretrained knowledge, it is crucial to retain and reuse knowledge from previously observed domains. We introduce Domain-Aware Spectral Code Retrieval, which estimates input distributions to detect domain shifts, and retrieves adapted singular values for rapid adaptation. Consequently, our method achieves state-of-the-art performance on various distribution-shift benchmarks under the TTA setting. In CTTA and Gradual CTTA, it further improves accuracy by 3.4 percentage points (pp) and 2.4 pp, respectively, while requiring 385 times fewer trainable parameters. Our code is available at https://github.com/baek85/IMSE.