Research map / Label-efficient and robust learning / Test-time adaptation
ZeroSiam: An Efficient Siamese for Test-Time Entropy Optimization without Collapse
Guohao Chen, Shuaicheng Niu, Deyu Chen, Jiahao Yang, Zitian Zhang, Mingkui Tan, Pengcheng Wu, Zhiqi Shen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add predictor with stop-gradient asymmetric alignment to Tent (ablation) | Tent · ImageNet-C level 5 online label shifts, ViT-Base | accuracy (%): 47.3 -> 64.1 | Improved |
| use a fixed random predictor instead of a learnable identity-initialized predictor (ablation) | ZeroSiam · ImageNet-C level 5 online label shifts, ViT-Base | accuracy (%): 64.1 -> 60.7 | Hurt |
| make the predictor deeper and nonlinear (FC+ReLU+FC) (ablation) | ZeroSiam · ImageNet-C level 5 online label shifts, ViT-Base | accuracy (%): 64.1 -> 64.0 | No clear effect |
| add ZeroSiam to EATA | EATA · ImageNet-C level 5 label shifts (ViT-B, R-50, ConvNeXt-T avg) | accuracy (%): 40.6 -> 49.0 | Improved |
| add ZeroSiam to DeYO | DeYO · ImageNet-C level 5 label shifts (ViT-B, R-50, ConvNeXt-T avg) | accuracy (%): 46.5 -> 53.9 | Improved |
| swap objective or divergence (KL, rKL, JS, MSE, sKL) in the alignment term (ablation) | ZeroSiam · ImageNet-C level 5 label shifts, ViT-Base, Tent loss | accuracy (%): 63.8 to 64.1 across divergences | No clear effect |
| add the asymmetric predictor to the SLR objective (ablation) | SLR · ImageNet-C level 5 label shifts, ViT-Base | accuracy (%): 44.0 -> 63.2 | Improved |
| set alpha to 1.0 versus larger values (ablation) | ZeroSiam · ImageNet-C level 5 online label shifts, ViT-Base | accuracy (%): 64.1 -> 62.4 (alpha 1.0 -> 2.5) | Hurt |
| vary predictor learning rate between 0.001 and 0.01 (ablation) | ZeroSiam · ImageNet-C level 5 label shifts, ViT-Base | accuracy (%): about 1% change | No clear effect |
| use ZeroSiam instead of DeYO on a weak model under label shifts | ViT-Small on noise corruptions · ImageNet-C level 5 label shifts | accuracy (%): 0.1 -> 26.5 | Improved |
| adapt only on misclassified blind-spot samples with ZeroSiam instead of prior methods | ImageNet-C level 5 batch size 1 · blind-spot subset | average accuracy (%): 29.0 -> 52.0 | Improved |
| add online entropy-based TTA (Tent, SAR) to Llama3.1-8B reasoning on CollegeMath | Llama3.1-8B-Instruct · CollegeMath | accuracy (%): 25.00 -> 24.17 (Tent), 25.00 -> 24.25 (SAR) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Tent — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- EAGer: Entropy-Aware GEneRation for Adaptive Inference-Time Scaling (ICML 2026 · similarity 0.87)
- Diversity Matters: Revisiting Test-Time Compute in Vision-Language Models (ICML 2026 · similarity 0.87)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.87)
- Ranked Entropy Minimization for Continual Test-Time Adaptation (ICML 2025 · similarity 0.87)
- Entropy is not Enough for Test-Time Adaptation: From the Perspective of Disentangled Factors (ICLR 2024 · similarity 0.86)
- A Closer Look at Model Collapse: From a Generalization-to-Memorization Perspective (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
