A Framework for Multiple-Instance Learning

By Oded Maron, Tomas Lozano-Perez. 1997

How researchers see this paper

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

Citations in the corpus

Contextual Action Recognition with R*CNN (2015, background): “Inspired by multiple-instance learning (MIL) [31, 21] and Latent SVM [9], if I is an image and r is a region in I containing the target person, we define the score of action α as score(α; I, r) = wαp ·φ(r; I) + max s∈R(r;I) wαs ·φ(s; I), (1) where φ(r; I) is a vector of features extracted from regio”
A Comprehensive Survey of Deep Learning for Image Captioning (2018, methodology): “Multiple instance learning (MIL) [96] is used to train the model for learning discriminative visual signatures of each word.”
Maximum Margin Multiple Instance Clustering With Applications to Image and Text Clustering (2011, methodology): “Maron and Lozano-Pérez proposed a method called diverse density (DD) [13], [14], which intends to find a concept point in the feature space that resembles positive instance most, and classify instances according to the distances between the instances and this concept point.”
Multiple-Instance Active Learning (2007, methodology): “in the context of drug activity prediction [3], and has since been applied to a wide variety of tasks including content-based image retrieval [1, 6, 8], text classification [1, 9], stock prediction [6], and protein family modeling [10].”
Weakly Supervised Semantic Segmentation with Convolutional Networks (2014, background): “This approach can be seen as an instance of Multiple Instance Learning (MIL) [12].”
AUTOMATED DISCOVERY OF OPTIONS IN REINFORCEMENT LEARNING (2003, methodology): “McGovern and Barto used the Diverse Density algorithm [16][17] to solve the multiple instance learning problem and find states that appear often on successful trajectories [18][19].”
MILCut: A Sweeping Line Multiple Instance Learning Paradigm for Interactive Image Segmentation (2014, background): “, DD [23], EM-DD [42], citation-kNN [37], MI-SVM and mi-SVM [4], and IQH [6], have been proposed to solve the multiple instance learning problem.”
Multi-instance learning with any hypothesis class (2011, methodology): “Diverse Density (Maron and Lozano-P´erez, 1998) and EM-DD (Zhang and Goldman, 2001) employ assumptions on the structure of the bags of instances.”
Multiple instance learning: A survey of problem characteristics and applications (2016, methodology): “In APR [3] the region is a hyper-rectangle, while in many others it is one, or a collection of, hyper-spheres/-ellipses [29, 31, 33, 88].”
Deep Learning in Medical Image Analysis (2017, methodology): “Since each slice may contain multiple organs (enclosed in the bounding boxes), their CNN was trained in multi-instance fashion (92), where the objective function in CNN was adapted to in a way that as long as one organ was correctly labeled, the corresponding slice was considered as correct.”
Discover and Learn New Objects from Documentaries (2017, unknown): “ cases, the supervised information is restricted to binary labels that indicate the absence/presence of object instances in the image, without their locations. Typically, a multiple instance learning [25] framework is adopted. Speciﬁcally, each im1Dataset and code are available at https://github.com”
Bayesian multiple instance learning: automatic feature selection and inductive transfer (2008, methodology): “(Maron & Lozano-Perez, 1998) proposed a framework called Diverse Density algorithm.”
Multiple Instance Learning: Algorithms and Applications (2008, methodology): “Having deﬁned this, we can again plug the instance probabilities into one of the max approximations, and then into the likelihood ([21] uses the NOR model).”
Mining Multi-label Data (2010, background): “Multiple-instance or multi-instance learning is a variation of supervised learning, where labels are assigned to bags of instances [66].”