Multi-instance learning by treating instances as non-I.I.D. samples

By Zhi-Hua Zhou, Yu-Yin Sun, Yu-Feng Li. 2008

493 citations recorded by the source corpus.

Previous studies on multi-instance learning typically treated instances in the bags as independently and identically distributed. The instances in a bag, however, are rarely independent in real tasks, and a better performance can be expected if the instances are treated in an non-i.i.d. way that exploits relations among instances. In this paper, we propose two simple yet effective methods. In the first method, we explicitly map every bag to an undirected graph and design a graph kernel for distinguishing the positive and negative bags. In the second method, we implicitly construct graphs by deriving affinity matrices and propose an efficient graph kernel considering the clique information. The effectiveness of the proposed methods are validated by experiments.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

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

In Computer Vision and Pattern Recognition

Citations in the corpus

Learning Theory for Distribution Regression (2014, background): “…consistency guarantees—for many other multi-instance problems in the literature including (i) regression (Ray and Page, 2001; Dooly et al., 2002; Zhou et al., 2009; Kwok and Cheung, 2007), (ii) clustering (Zhang and Zhou, 2009; Zhang et al., 2009, 2011; Chen and Wu, 2012), (iii) ranking…”
Attention-based Deep Multiple Instance Learning (2018, neutral): “[1] (Andrews et al., 2003), [2] (G¨artner et al., 2002), [3] (Zhang & Goldman, 2002) [4] (Zhou et al., 2009) [5] (Wei et al., 2017)”
Multiple instance classification: Review, taxonomy and comparative study (2013, background): “Along the same line, Zhou et al [28] proposes another kernel function K ( X , Y ).”
Revisiting multiple instance neural networks (2016, methodology): “Dataset MI-Kernel [25] miGraph [26] miFV [27] mi-Net MI-Net MI-Net with DS MI-Net with RC”
A survey on instance selection for active learning (2012, background): “– Exploiting on feature correlation: Usually, a similarity measurement [62] or a correlation matrix [51] on features is utilized to compare pairwise similarities of instances, so the informativeness of an instance is weighted by average similarity on its neighbors.”
Multi-Instance Learning With Emerging Novel Class (2021, methodology): “Many existing MIL works [1], [5], [6], [7], [8], [9] solve the binary MIL classiﬁcation problem.”
On classification with bags, groups and sets (2014, background): “In advertising, the proportions are related to customers that would buy a product only on discount, and customers that would buy a product in any circumstances.”
Correlated Differential Privacy: Hiding Information in Non-IID Data Set (2015, background): “However, if some records are correlated with each other, traditional differential privacy may under-estimate the privacy risk [2].”
Non-I.I.D. Multi-Instance Dimensionality Reduction by Learning a Maximum Bag Margin Subspace (2010, methodology): “In this section, we introduce our MidLABS algorithm, which takes advantage of both discriminant information of bags and geometrical structures of instances in each bag.”
A brief introduction to weakly supervised learning (2018, background): “samples, and based on this insight, some effective algorithms have been developed [47].”
Multiple instance learning: A survey of problem characteristics and applications (2016, methodology): “To deal with similar instances, miGraph [10] builds a graph per bag and groups similar instances together to adjust their relative importance based on the group size.”
Exploit Bounding Box Annotations for Multi-Label Object Recognition (2015, methodology): “Over the years, many multi-instance learning algorithms have been proposed, including miBoosting [29], miSVM [1], MILES [7] and miGraph [33].”
Aspect-Controllable Opinion Summarization (2021, background): “, 1997; Maron and Ratan, 1998) or jointly for instances and bags (Zhou et al., 2009; Wei et al., 2014; Kotzias et al., 2015; Xu and Lapata, 2019; Angelidis and Lapata, 2018a).”
Multiple instance learning for soft bags via top instances (2015, methodology): “Various solutions and applications have since been proposed, based on both generative [16, 4] and discriminant [1, 11, 25, 6, 5, 3, 31, 15, 30, 8, 12] formulations of the problem.”
A review of multi-instance learning assumptions (2010, methodology): “In (Zhou et al., 2009), the WIis are computed using the Gaussian distance, consistently with the Gaussian RBF kernel used for knode.”