arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2308.07477cs.CVeess.IV
收藏

概率化 MIMO U-Net:面向像素级回归的高效且准确的不确定性估计

Probabilistic MIMO U-Net: Efficient and Accurate Uncertainty Estimation for Pixel-wise Regression

University of the Bundeswehr Munich(慕尼黑联邦国防军大学)

Anton Baumann, Thomas Roßberg, Michael Schmitt

AI总结:

本文将 MIMO 框架适配到 U-Net 像素级回归,通过单模型内训练多个子网络并同步其性能,实现兼具准确率、校准性、分布外检测与效率的不确定性估计。

摘要:

机器学习中的不确定性估计对于提升预测模型的可靠性和可解释性至关重要,尤其是在高风险的真实世界场景中。尽管已有许多方法可用,但它们往往需要在不确定性估计质量与计算效率之间进行权衡。为应对这一挑战,我们提出了 Multiple-Input Multiple-Output(MIMO,多输入多输出)框架的一种适配版本——该方法利用深度神经网络的过参数化特性——用于像素级回归任务。我们的 MIMO 变体将该方法的适用性从简单的图像分类扩展到更广泛的计算机视觉领域。为此,我们改造了 U-Net 架构,以在单个模型中训练多个子网络,从而利用深度神经网络中的过参数化。此外,我们还提出了一种在 MIMO 框架内同步子网络性能的新流程。我们在两个正交数据集上对所得 MIMO U-Net 进行了全面评估,结果表明其准确率与现有模型相当,在分布内数据上具有更优的校准效果,具备稳健的分布外检测能力,并且在参数量和推理时间方面有显著改善。代码可在 github.com/antonbaumann/MIMO-Unet 获取。

英文摘要:

Uncertainty estimation in machine learning is paramount for enhancing the reliability and interpretability of predictive models, especially in high-stakes real-world scenarios. Despite the availability of numerous methods, they often pose a trade-off between the quality of uncertainty estimation and computational efficiency. Addressing this challenge, we present an adaptation of the Multiple-Input Multiple-Output (MIMO) framework -- an approach exploiting the overparameterization of deep neural networks -- for pixel-wise regression tasks. Our MIMO variant expands the applicability of the approach from simple image classification to broader computer vision domains. For that purpose, we adapted the U-Net architecture to train multiple subnetworks within a single model, harnessing the overparameterization in deep neural networks. Additionally, we introduce a novel procedure for synchronizing subnetwork performance within the MIMO framework. Our comprehensive evaluations of the resulting MIMO U-Net on two orthogonal datasets demonstrate comparable accuracy to existing models, superior calibration on in-distribution data, robust out-of-distribution detection capabilities, and considerable improvements in parameter size and inference time. Code available at github.com/antonbaumann/MIMO-Unet

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。

↑