EVA-0: Test-Time Model Evolution with Only Two Forward Passes per Sample
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866913143601496064 |
|---|---|
| author | Chen, Guohao Niu, Shuaicheng Li, Geng Zhang, Yunbei Shan, Shilin Miao, Chunyan Yang, Jianfei |
| author_facet | Chen, Guohao Niu, Shuaicheng Li, Geng Zhang, Yunbei Shan, Shilin Miao, Chunyan Yang, Jianfei |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_18867 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | EVA-0: Test-Time Model Evolution with Only Two Forward Passes per Sample Chen, Guohao Niu, Shuaicheng Li, Geng Zhang, Yunbei Shan, Shilin Miao, Chunyan Yang, Jianfei Machine Learning Artificial Intelligence 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. |
| title | EVA-0: Test-Time Model Evolution with Only Two Forward Passes per Sample |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2605.18867 |