EVA-0: Test-Time Model Evolution with Only Two Forward Passes per Sample
About
Test-time model evolution offers a promising way for deployed models to improve from unlabeled test-time experience, yet most existing methods depend on backpropagation (BP), which incurs substantial memory overhead and makes them difficult to deploy on edge devices, quantized models, specialized accelerators, or black-box models. In this work, we study test-time model evolution under a strict two-forward budget, a setting that pushes adaptation toward highly efficient real-world deployment. We reveal three key obstacles in zeroth-order test-time optimization: susceptibility to shortcut solutions, uncontrolled weight drift, and ineffective update direction estimation. To overcome them, we propose EVA-0, a minimal zeroth-order adaptation framework that: 1) keeps the loss scale-invariant to prevent shortcut solutions; 2) devises an anchor-guided optimization strategy to alleviate weight drift; 3) uses sample-wise symmetric two-sided perturbation for update direction estimation and inference. EVA-0 requires no BP and performs both inference and adaptation within only two forward passes per sample. Results on ImageNet-C & ViT-Base show that EVA-0 outperforms both BP-based DeYO and BP-free FOA, while achieving a 14x speed-up over FOA. Code will be released.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Image Classification | ImageNet-C level 5 | -- | 110 | |
| Online Continual Test-Time Adaptation | ImageNet-C Severity 5 (test) | Accuracy (Gaussian Noise, ImageNet-C S5)61.8 | 47 | |
| Image Classification | ImageNet-C Severity 5 | Error Rate (Gaussian)38.2 | 43 | |
| Image Classification | ImageNet-C severity level 5 v1 (test) | Accuracy (Gaussian)61.8 | 22 | |
| Test-time adaptation | ImageNet-C level 5 | Accuracy66.1 | 10 |