(b) Accuracy-Runtime Trade-off Figure 2(b) Accuracy-Runtime Trade-off Figure 2. (a) Controlling shrinkage strength with anchor weight α of state-of-the-art method StatA. Such static modeling is suboptimal in accuracy and robustness. (b) Accuracy-Runtime Tradeoff. MOON enables effective and efficient adaptation.这张图/表用于判断 Von Mises-Fisher Mixture Model with 的实验收益来自哪里。重点看替换、消融或跨模型设置下趋势是否一致，而不是只看单个最高分。(b) Online adaptation under non-i.i.d(b) Online adaptation under non-i.i.d. data streams. Figure 1. Performance comparison on two realistic settings. Existing transductive or online TTA methods suffer from performance degradation or even collapse, while our proposed MOON consistently enhances VLM prediction and outperforms state-of-the-art baselines.这张图/表用于判断 Von Mises-Fisher Mixture Model with 的实验收益来自哪里。重点看替换、消融或跨模型设置下趋势是否一致，而不是只看单个最高分。
