Support Vector Machines for Multiple-Instance Learning

By Stuart Andrews, Ioannis Tsochantaridis, Thomas Hofmann. 2002

1,662 citations recorded by the source corpus.

How researchers see this paper

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

In Statistical Learning Theory

Citations in the corpus

End-to-End Learning of Visual Representations From Uncurated Instructional Videos (2019, neutral): “The max-pool based approach [5, 8, 58] (Max+NCE) only optimizes over the clip-caption pair with the highest similarity score among the positive candidates.”
Fully Convolutional Multi-Class Multiple Instance Learning (2014, methodology): “Most MIL problems are framed as max-margin learning (Andrews et al., 2002; Felzenszwalb et al., 2010), while other approaches use boosting (Ali & Saenko, 2014) or Noisy-OR models (Heckerman, 2013).”
Constrained Convolutional Neural Networks for Weakly Supervised Segmentation (2015, methodology): “This foreground constraint is very similar to the commonly used multiple instance learning (MIL) paradigm, where at least one pixel is constrained to be positive [2, 16, 25, 26].”
Attention-based Deep Multiple Instance Learning (2018, neutral): “, 2015b), embedding instances to a compact low-dimensional representation that is further fed to a bag-level classifier (Andrews et al., 2003; Chen et al., 2006), and combining responses of an instance-level classifier (Ramon & De Raedt, 2000; Raykar et al.”
Modeling Context Between Objects for Referring Expression Understanding (2016, neutral): “Our two formulations of the training objective functions are conceptually similar to MISVM and mi-SVM [9].”
Beyond Nouns: Exploiting Prepositions and Comparative Adjectives for Learning Visual Classifiers (2008, methodology): “Many learning approaches have been used for annotating images which include translation models [5], statistical models [2,6], classification approaches [ 7 ,8,9] and relevance language models [10,11].”
A Deep Neural Framework for Continuous Sign Language Recognition by Iterative Training (2019, methodology): “Pfister et al. [27] use subtitle text, lip and hand motion cues to select candidate temporal windows, and these candidates are further refined using MI-SVM [32].”
A survey on face detection in the wild: Past, present and future (2015, methodology): “The above methodology, called latent-SVM, is similar to the multi-instance SVMs proposed in [174].”
Hopfield Networks is All You Need (2020, background): “, 2014), and the Elephant, Fox, Tiger datasets (Andrews et al., 2003).”
Multiple instance ranking (2008, methodology): “Raedt, 2000), and generalisations of support vector machines ( Andrews et al., 2003; Mangasarian & Wild, 2008).”
Audio event and scene recognition: A unified approach using strongly and weakly labeled data (2016, methodology): “[17] used SVM (miSVM) [22] and neural network (BPMIL) [23] based MIL methods to show that AED with weak labels can be successfully done.”
Spot On: Action Localization from Pointly-Supervised Proposals (2016, methodology): “The common tactic in the literature is to model this as a Multiple Instance Learning (MIL) problem [8,9,10], where positive images contain at least one”
Harvesting Mid-level Visual Concepts from Large-Scale Internet Images (2013, methodology): “In [1], the relationship between the bag level labels and the instance level labels is formulated as a set of linear constraints.”
Weakly Supervised Object Localization with Latent Category Learning (2014, neutral): “4 MIL-SVM [2] 37.”
MILES: Multiple-Instance Learning via Embedded Instance Selection (2006, methodology): “[2] formulated MIL as a mixed integer quadratic program.”
Seed, Expand and Constrain: Three Principles for Weakly-Supervised Image Segmentation (2016, background): “Variants of multiple instance learning [1] train with a per-image loss function, while internally maintaining a spatial representation of the image that can be used to produce segmentation masks [37,38,39].”
Multiple instance learning for soft bags via top instances (2015, methodology): “The positive source is the distribution of the target concept (e.g., image patches of “flower”), the negative source the distribution of background clutter (e.g., image patches of everything else).”