Multi-instance multi-label learning

By Zhi-Hua Zhou, Min-Ling Zhang, Sheng-Jun Huang, Yu-Feng Li. Artificial Intelligence, 2008

431 citations recorded by the source corpus.

In this paper, we propose the MIML (Multi-Instance Multi-Label learning) framework where an example is described by multiple instances and associated with multiple class labels. Compared to traditional learning frameworks, the MIML framework is more convenient and natural for representing complicated objects which have multiple semantic meanings. To learn from MIML examples, we propose the MimlBoost and MimlSvm algorithms based on a simple degeneration strategy, and experiments show that solving problems involving complicated objects with multiple semantic meanings in the MIML framework can lead to good performance. Considering that the degeneration process may lose information, we propose the D-MimlSvm algorithm which tackles MIML problems directly in a regularization framework. Moreover, we show that even when we do not have access to the real objects and thus cannot capture more information from real objects by using the MIML representation, MIML is still useful. We propose the InsDif and SubCod algorithms. InsDif works by transforming single-instances into the MIML representation for learning, while SubCod works by transforming single-label examples into the MIML representation for learning. Experiments show that in some tasks they are able to achieve better performance than learning the single-instances or single-label examples directly.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Signal Processing

View Paper: https://www.semanticscholar.org/paper/fae4506bd003b597fab3291e89fd25b6e3f56bad

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Deep Multi-instance Networks with Sparse Label Assignment for Whole Mammogram Classification (2016, background): “extended multi-instance learning to multi-class classification problems [23].”
A Review on Multi-Label Learning Algorithms (2014, background): “In contrast to multi-label learning which models the object’s ambiguities (complicated semantics) in output (label) space, multi-instance learning can be viewed as modeling the object’s ambiguities in input (instance) space [113].”
Relation Extraction with Multi-instance Multi-label Convolutional Neural Networks (2016, methodology): “This multi-instance multi-label (MIML) learning scenario was formulated in Zhou et al. (2012), and get widely employed in various tasks (Zha et al., 2008; Zhou and Zhang, 2006; Li et al., 2012).”
Multiple instance learning: A survey of problem characteristics and applications (2016, background): “This idea has been the object of several publications [37, 38, 39].”
Transductive Multilabel Learning via Label Set Propagation (2013, background): “Zhou et al. [39], [40] proposed the MIML framework which deals with multilabel examples each is represented as a set of instances.”
On classification with bags, groups and sets (2014, neutral): “Examples can be found in [62, 51].”
LEAP: Learning to Prescribe Effective and Safe Treatment Combinations for Multimorbidity (2017, methodology): “We formulate the treatment recommendation as a Multi-Instance Multi-Label (MIML) learning problem [42], where training samples are provided as a set of instances (i.e., disease conditions) and the associated label set (i.e., medications).”
Multi-Label Learning with Global and Local Label Correlation (2017, background): “Besides, it is also interesting to extend our work to a multi-instance multi-label setting [53], where each object is represented by several instances, and structures among them can be taken into account.”
A Tutorial on Multilabel Learning (2015, background): “Because in many MLL problems, labels can be related only to different parts of the object; thus, developing learners on a MIML setting may reduce noise and improve performance [Zhou et al. 2012b].”