PulseAugur
实时 13:13:14
English(EN) Entropy Minimization without Model Collapse: Mitigating Prediction Bias in Medical Imaging

新方法对抗医学影像AI中的预测偏差

研究人员发现了一种测试时自适应方法中的关键失效模式,称为模型坍塌,在这种模式下,类别簇会合并并导致预测偏差。他们提出了一种新的目标函数——分布偏移偏差减少(DSBR),通过确保损失函数中各类别贡献的平衡来抵消这一问题。在医学影像数据集和ImageNet-C上的实验表明,DSBR能有效稳定自适应过程,防止模型坍塌,并取得具有竞争力或更优的性能。 AI

影响 减轻了测试时自适应中的一种关键失效模式,有望提高AI模型在医学影像等关键应用中的可靠性。

排序理由 该集群包含一篇学术论文,详细介绍了一种缓解AI模型自适应中特定问题的新方法。

在 arXiv cs.LG 阅读 →

AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →

报道来源 [2]

  1. arXiv cs.LG TIER_1 English(EN) · Tim Nielen, Sameer Ambekar, Johannes Kiechle, Daniel M. Lang, Julia A. Schnabel ·

    Entropy Minimization without Model Collapse: Mitigating Prediction Bias in Medical Imaging

    arXiv:2606.02339v1 Announce Type: new Abstract: Entropy minimization (EM) is the dominant objective for test-time adaptation, yet its failure mode, model collapse, remains poorly understood. In this work, we show that distribution shifts can cause feature clusters corresponding t…

  2. arXiv cs.LG TIER_1 English(EN) · Julia A. Schnabel ·

    Entropy Minimization without Model Collapse: Mitigating Prediction Bias in Medical Imaging

    Entropy minimization (EM) is the dominant objective for test-time adaptation, yet its failure mode, model collapse, remains poorly understood. In this work, we show that distribution shifts can cause feature clusters corresponding to distinct classes in the model's representation…