EVA-0: Test-Time Model Evolution with Only Two Forward Passes per Sample
The paper presents EVA-0, a novel framework for test-time model evolution that operates with only two forward passes per sample. This approach addresses the challenges of existing methods that rely on backpropagation, which can be resource-intensive and difficult to implement on edge devices. EVA-0 demonstrates improved performance and efficiency, significantly speeding up the adaptation process in machine learning models.
- ▪EVA-0 allows for model adaptation without backpropagation, reducing memory overhead.
- ▪The framework addresses issues such as shortcut solutions and weight drift.
- ▪EVA-0 achieves a 14x speed-up over previous methods while maintaining performance.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Machine Learning arXiv:2605.18867 (cs) [Submitted on 15 May 2026] Title:EVA-0: Test-Time Model Evolution with Only Two Forward Passes per Sample Authors:Guohao Chen, Shuaicheng Niu, Geng Li, Yunbei Zhang, Shilin Shan, Chunyan Miao, Jianfei Yang View a PDF of the paper titled EVA-0: Test-Time Model Evolution with Only Two Forward Passes per Sample, by Guohao Chen and 6 other authors View PDF HTML (experimental) Abstract: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.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
