arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2405.08766cs.LGcs.CV

用于分布外检测的基于能量的 Hopfield Boosting

Energy-based Hopfield Boosting for Out-of-Distribution Detection

Software Competence Center Hagenberg GmbH(哈根伯格软件能力中心有限公司);Silicon Austria Labs(奥地利硅谷实验室);Helmholtz Centre for Environmental Research–UFZ(亥姆霍兹环境研究中心)

Claus Hofmann, Simon Schmid, Bernhard Lehner, Daniel Klotz, Sepp Hochreiter

AI总结:

提出基于现代 Hopfield 能量的 Hopfield Boosting,通过加权靠近决策边界的难分辅助异常样本,显著提升异常暴露式 OOD 检测性能。

摘要:

在现实世界中部署机器学习模型时,分布外(OOD)检测至关重要。异常暴露方法会在训练过程中引入辅助异常数据;与没有先进训练策略的方法相比,这类方法能够显著提升 OOD 检测性能。我们提出 Hopfield Boosting,这是一种利用现代 Hopfield 能量(MHE)来锐化分布内数据与 OOD 数据之间决策边界的提升方法。Hopfield Boosting 促使模型聚焦于难以区分的辅助异常样本,这些样本靠近分布内数据与辅助异常数据之间的决策边界。我们的方法在采用异常暴露的 OOD 检测中达到了新的最优水平,在 CIFAR-10 上将 FPR95 指标从 2.28 改善至 0.92,在 CIFAR-100 上从 11.76 改善至 7.94。

英文摘要:

Out-of-distribution (OOD) detection is critical when deploying machine learning models in the real world. Outlier exposure methods, which incorporate auxiliary outlier data in the training process, can drastically improve OOD detection performance compared to approaches without advanced training strategies. We introduce Hopfield Boosting, a boosting approach, which leverages modern Hopfield energy (MHE) to sharpen the decision boundary between the in-distribution and OOD data. Hopfield Boosting encourages the model to concentrate on hard-to-distinguish auxiliary outlier examples that lie close to the decision boundary between in-distribution and auxiliary outlier data. Our method achieves a new state-of-the-art in OOD detection with outlier exposure, improving the FPR95 metric from 2.28 to 0.92 on CIFAR-10 and from 11.76 to 7.94 on CIFAR-100.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑