论文阅读 [TPAMI-2022] Large-Scale Nonlinear AUC Maximization via Triply Stochastic Gradients

论文阅读 [TPAMI-2022] Large-Scale Nonlinear AUC Maximization via Triply Stochastic Gradients

论文搜索(studyai.com)

搜索论文: Large-Scale Nonlinear AUC Maximization via Triply Stochastic Gradients

搜索论文: http://www.studyai.com/search/whole-site/?q=Large-Scale+Nonlinear+AUC+Maximization+via+Triply+Stochastic+Gradients

关键字(Keywords)

Kernel; Stochastic processes; Training; Approximation algorithms; Optimization; Measurement; Learning systems; AUC maximization; random fourier features; kernel methods

机器学习; 运筹与优化

梯度下降; 类(样本)不平衡

摘要(Abstract)

Learning to improve AUC performance for imbalanced data is an important machine learning research problem.

学习提高不平衡数据的AUC性能是一个重要的机器学习研究问题。.

Most methods of AUC maximization assume that the model function is linear in the original feature space.

大多数AUC最大化方法都假设模型函数在原始特征空间中是线性的。.

However, this assumption is not suitable for nonlinear separable problems.

然而，这种假设不适用于非线性可分问题。.

Although there have been some nonlinear methods of AUC maximization, scaling up nonlinear AUC maximization is still an open question.

虽然已经有一些非线性的AUC最大化方法，但扩大非线性AUC最大化仍然是一个悬而未决的问题。.

To address this challenging problem, in this paper, we propose a novel large-scale nonlinear AUC maximization method (named as TSAM) based on the triply stochastic gradient descents.

为了解决这个具有挑战性的问题，本文提出了一种基于三重随机梯度下降的大规模非线性AUC最大化方法（简称TSAM）。.

Specifically, we first use the random Fourier feature to approximate the kernel function.

具体来说，我们首先使用随机傅立叶特征来近似核函数。.

After that, we use the triply stochastic gradients w.r.t.

然后，我们使用三重随机梯度w.r.t。.

the pairwise loss and random feature to iteratively update the solution.

两两损失和随机特征迭代更新解决方案。.

Finally, we prove that TSAM converges to the optimal solution with the rate of $ \mathcal {O}(1/t)$O(1/t) after $t$t iterations.

最后，我们证明了TSAM在经过$t$t迭代后以$\mathcal{O}（1/t）$O（1/t）的速率收敛到最优解。.

Experimental results on a variety of benchmark datasets not only confirm the scalability of TSAM, but also show a significant reduction of computational time compared with existing batch learning algorithms, while retaining the similar generalization performance…

在各种基准数据集上的实验结果不仅证实了TSAM的可扩展性，而且与现有的批学习算法相比，计算时间显著减少，同时保持了相似的泛化性能。。.

作者(Authors)

[‘Zhiyuan Dang’, ‘Xiang Li’, ‘Bin Gu’, ‘Cheng Deng’, ‘Heng Huang’]