A Convex Method for Locating Regions of Interest with Multi-instance Learning

Li, Yufeng; Kwok, James T.; Tsang, Ivor W.; Zhou, Zhi‐Hua

doi:10.1007/978-3-642-04174-7_2

Cited by 77 publications

(80 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We briefly discuss two interesting approaches. In (Li et al, 2009), the CBIR problem is particularly considered where the regions of interest can be seen as witnesses or key instances in positive bags. They formed a convex optimization problem iteratively by finding violated key instances and combining them via multiple kernel learning.…”

Section: Related Workmentioning

confidence: 99%

Multiple instance learning via Gaussian processes

Kim

Torre

2013

Data Min Knowl Disc

View full text Add to dashboard Cite

Multiple instance learning (MIL) is a binary classification problem with loosely supervised data where a class label is assigned only to a bag of instances indicating presence/absence of positive instances. In this paper we introduce a novel MIL algorithm using Gaussian processes (GP). The bag labeling protocol of the MIL can be effectively modeled by the sigmoid likelihood through the max function over GP latent variables. As the non-continuous max function makes exact GP inference and learning infeasible, we propose two approximations: the soft-max approximation and the introduction of witness indicator variables. Compared to the state-of-the-art MIL approaches, especially those based on the Support Vector Machine (SVM), our model enjoys two most crucial benefits: (i) the kernel parameters can be learned in a principled manner, thus avoiding grid search and being able to exploit a variety of kernel families with complex forms, and (ii) the efficient gradient search for kernel parameter learning effectively leads to feature selection to extract most relevant features while discarding noise. We demonstrate that our approaches attain superior or comparable performance to existing methods on several real-world MIL datasets including large-scale content-based image retrieval problems.

show abstract

Section: Related Workmentioning

confidence: 99%

Multiple instance learning via Gaussian processes

Kim

Torre

2013

Data Min Knowl Disc

View full text Add to dashboard Cite

show abstract

“…The instances classified as positive are the witnesses. KI-SVM [4] also locates ROI by finding the key instance (i.e. witness) in bags using multiple kernel learning.…”

Section: Witness Identification In Mil Methodsmentioning

confidence: 99%

“…However, some of these methods, like MILES [5] and Citation-kNN [15], can be adapted for the task. In contrast, instance-based MIL methods like axis parallel rectangle (APR) [2], mi-SVM, MI-SVM [3] and KI-SVM [4] infer bag labels based on individual instance classification, and thus can be used directly for witness identification. Although these methods can achieve a high level of performance in specific situations, they often perform poorly when the proportion of positive instances in positive bags, hereafter called the witness rate (WR), is low.…”

Section: Introductionmentioning

confidence: 99%

Witness identification in multiple instance learning using random subspaces

Carbonneau

Granger

Gagnon

2016

2016 23rd International Conference on Pattern Recognition (ICPR)

View full text Add to dashboard Cite

Abstract-Multiple instance learning (MIL) is a form of weakly-supervised learning where instances are organized in bags. A label is provided for bags, but not for instances. MIL literature typically focuses on the classification of bags seen as one object, or as a combination of their instances. In both cases, performance is generally measured using labels assigned to entire bags. In this paper, the MIL problem is formulated as a knowledge discovery task for which algorithms seek to discover the witnesses (i.e. identifying positive instances), using the weak supervision provided by bag labels. Some MIL methods are suitable for instance classification, but perform poorly in application where the witness rate is low, or when the positive class distribution is multimodal. A new method that clusters data projected in random subspaces is proposed to perform witness identification in these adverse settings. The proposed method is assessed on MIL data sets from three application domains, and compared to 7 reference MIL algorithms for the witness identification task. The proposed algorithm constantly ranks among the best methods in all experiments, while all other methods perform unevenly across data sets.

show abstract

“…That is, we report the classification accuracies averaged over 10 runs where the parameter selection is carried our by using 10-fold cross validation. Our results are shown in Table 2 together with those of 12 other MIL algorithms in the literature [13,8,4,5,10,12,14,1,24]. All reported results are also based on 10-fold CV averaged over 10 runs 4 , with the exception of MIForest, which is over 5 runs, and MILIS and MIO, which are over 15 runs.…”

Section: Benchmark Data Setsmentioning

confidence: 97%

“…However, it is more important to note that it gives the best performance among the instance-selection based MIL approaches. [12] 88.3 87.7 n/a n/a n/a Ins-KI-SVM [14] 84 Table 2. Classification accuracies of various MIL algorithms on standard benchmark data sets.…”

Section: Benchmark Data Setsmentioning

confidence: 99%

Multiple-Instance Learning with Instance Selection via Dominant Sets

Erdem

2011

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. Multiple-instance learning (MIL) deals with learning under ambiguity, in which patterns to be classified are described by bags of instances. There has been a growing interest in the design and use of MIL algorithms as it provides a natural framework to solve a wide variety of pattern recognition problems. In this paper, we address MIL from a view that transforms the problem into a standard supervised learning problem via instance selection. The novelty of the proposed approach comes from its selection strategy to identify the most representative examples in the positive and negative training bags, which is based on an effective pairwise clustering algorithm referred to as dominant sets. Experimental results on both standard benchmark data sets and on multi-class image classification problems show that the proposed approach is not only highly competitive with state-of-the-art MIL algorithms but also very robust to outliers and noise.

show abstract

A Convex Method for Locating Regions of Interest with Multi-instance Learning

Cited by 77 publications

References 15 publications

Multiple instance learning via Gaussian processes

Multiple instance learning via Gaussian processes

Witness identification in multiple instance learning using random subspaces

Multiple-Instance Learning with Instance Selection via Dominant Sets

Contact Info

Product

Resources

About