论文阅读 [TPAMI-2022] Incomplete Label Multiple Instance Multiple Label Learning

论文阅读 [TPAMI-2022] Incomplete Label Multiple Instance Multiple Label Learning

论文搜索(studyai.com)

搜索论文: Incomplete Label Multiple Instance Multiple Label Learning

搜索论文: http://www.studyai.com/search/whole-site/?q=Incomplete+Label+Multiple+Instance+Multiple+Label+Learning

关键字(Keywords)

Labeling; Training; Phase locked loops; Birds; Numerical models; Graphical models; Standards; Incomplete-label learning; learning with missing labels; multiple-instance multiple-label learning; multi-instance multi-label learning; maximum likelihood; marginal maximum

机器学习

监督学习; 半监督学习; 多标签学习; 图网络; 极大似然估计

摘要(Abstract)

With increasing data volumes, the bottleneck in obtaining data for training a given learning task is the cost of manually labeling instances within the data.

随着数据量的增加，获取用于训练给定学习任务的数据的瓶颈是在数据中手动标记实例的成本。.

To alleviate this issue, various reduced label settings have been considered including semi-supervised learning, partial- or incomplete-label learning, multiple-instance learning, and active learning.

为了缓解这个问题，人们考虑了各种简化标签设置，包括半监督学习、部分或不完全标签学习、多实例学习和主动学习。.

Here, we focus on multiple-instance multiple-label learning with missing bag labels.

在这里，我们将重点关注丢失行李标签的多实例多标签学习。.

Little research has been done for this challenging yet potentially powerful variant of incomplete supervision learning.

对于这种具有挑战性但潜在强大的不完全监督学习变体，几乎没有进行过研究。.

We introduce a novel discriminative probabilistic model for missing labels in multiple-instance multiple-label learning.

在多实例多标签学习中，我们提出了一种新的缺失标签判别概率模型。.

To address inference challenges, we introduce an efficient implementation of the EM algorithm for the model.

为了解决推理难题，我们为模型引入了EM算法的高效实现。.

Additionally, we consider an alternative inference approach that relies on maximizing the label-wise marginal likelihood of the proposed model instead of the joint likelihood.

此外，我们考虑另一种推理方法，依赖于最大化标签的边际似然的建议模型，而不是联合似然。.

Numerical experiments on benchmark datasets illustrate the robustness of the proposed approach.

在基准数据集上的数值实验表明了该方法的鲁棒性。.

In particular, comparison to state-of-the-art methods shows that our approach introduces a significantly smaller decrease in performance when the proportion of missing labels is increased…

特别是，与最先进的方法相比，当丢失标签的比例增加时，我们的方法带来的性能下降要小得多。。.

作者(Authors)

[‘Tam Nguyen’, ‘Raviv Raich’]