Sample selection approaches are popular in robust learning from noisy labels.
However, how to properly control the selection process so that deep networks
can benefit from the memorization effect is a hard problem. In this paper,
motivated by the success of automated machine learning (AutoML), we model this
issue as a function approximation problem. Specifically, we design a
domain-specific search space based on general patterns of the memorization
effect and propose a novel Newton algorithm to solve the bi-level optimization
problem efficiently. We further provide theoretical analysis of the algorithm,
which ensures a good approximation to critical points. Experiments are
performed on benchmark data sets. Results demonstrate that the proposed method
is much better than the state-of-the-art noisy-label-learning approaches, and
also much more efficient than existing AutoML algorithms.

本文利用机器学习自动化的方法，设计出一个基于记忆效应的特定领域的搜索空间，并且提出了一种新的牛顿算法，能够有效地解决二级优化问题，在多个数据集上进行的实验结果表明，该算法比现有嘈杂标签学习方法更好，并且比现有自动 ML 算法更加高效。

利用记忆效应学习受损标签

Searching to Exploit Memorization Effect in Learning from Corrupted  Labels

We introduce and prove convergence of a damped Newton algorithm to
approximate solutions of the semi-discrete optimal transport problem with
storage fees, corresponding to a problem with hard capacity constraints. This
is a variant of the optimal transport problem arising in queue penalization
problems, and has applications to data clustering. Our result is novel as it is
the first numerical method with proven convergence for this variant problem;
additionally the algorithm applies to the classical semi-discrete optimal
transport problem but does not require any connectedness assumptions on the
support of the source measure, in contrast with existing results. Furthermore
we find some stability results of the associated Laguerre cells. All of our
results come with quantitative rates. We also present some numerical examples.

本文介绍和证明了一种阻尼牛顿算法的收敛性，以近似解决存储费用、硬容量约束下的半离散最优输运问题，并探讨了其在队列罚函数问题、数据聚类等方面的应用，是第一种基于数值方法并具有收敛性的解决该变体问题的算法，同时不需要任何关于源测度支集的连通性假设并提出相关的拉格尔细胞稳定性结果，所有结果均附带定量的速率分析和数值例子。