Learning structural SVMs with latent variables

By Chun-Nam John Yu, Thorsten Joachims. 2009

735 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 Artificial Intelligence

Citations in the corpus

Practical Bayesian Optimization of Machine Learning Algorithms (2012, methodology): “Setting the hyperparameters, such as the regularisation term, C , of structured SVMs remains a challenge and these are typically set through a time consuming grid search procedure as is done in Miller et al. (2012) and Yu and Joachims (2009).”
Constrained Convolutional Neural Networks for Weakly Supervised Segmentation (2015, methodology): “MISVM [2] and LSVM [10, 35] are some of the classic methods in this paradigm, which are extension of SVMs to MIL scenario.”
Efficient Non-greedy Optimization of Decision Trees (2015, methodology): “Our upper bound is inspired by the formulation of structured prediction with latent variables [26].”
Learning Answer-Entailing Structures for Machine Comprehension (2015, methodology): “We use a convex upper bound of the loss function (Yu and Joachims, 2009) to rewrite the objective: This problem can be solved using Concave-Convex Programming (Yuille and Rangarajan, 2003) with the cutting plane algorithm for structural SVM (Finley and Joachims, 2008).”
CRAFT Objects from Images (2016, background): “Structural SVM [33, 18] and its latent version [37] are widely used when the problem has a structural loss.”
A Hierarchical Pose-Based Approach to Complex Action Understanding Using Dictionaries of Actionlets and Motion Poselets (2016, methodology): “We formulate learning the model parameters as a Latent Structural SVM problem [39], with latent variables for motion poselets z and actionlets v.”
Learning Attitudes and Attributes from Multi-aspect Reviews (2012, background): “It is certainly possible that by using latent-variable structured learning techniques our fully-supervised algorithm could be extended to make use of unlabeled data [35].”
Learning latent structure for activity recognition (2014, methodology): “Following [16] and [17], we substitute the loss function in (7) by the margin rescaling surrogate which serves as an upper-bound of the loss function.”
Discriminative figure-centric models for joint action localization and recognition (2011, methodology): “Inspired by the latent SVM [8, 25], we use the following scoring function to measure the compatibility between a video I, an action label y and the configurations of bounding boxes L that localize the person performing the action in each frame of the video: fθ(L, y, I) = maxz θ>Φ(z, L, y, I), where ”
Strong supervision from weak annotation: Interactive training of deformable part models (2011, methodology): “In this section, we present an algorithm that computes and displays in realtime the maximum likelihood location of a deformable part model as the user drags different parts with the mouse.”
Learning to Detect Partially Overlapping Instances (2013, methodology): “To handle this, we follow a conventional way [25] and add the “ground truth” configuration for each image into the optimization (3) as a latent variable h ∈ Hj (where Hj denotes the set of all labelings with the zero IC-loss).”
A cross-benchmark comparison of 87 learning to rank methods (2015, neutral): “The study reported evaluation results in a different metric than the metrics chosen for this meta-analysis (Kersting & Xu, 2009; Mohan, Chen, & Weinberger, 2011; Pahikkala, Tsivtsivadze, Airola, Järvinen, & Boberg, 2009; Thuy, Vien, Viet, & Chung, 2009; Yu & Joachims, 2009).”
Learning from Explicit and Implicit Supervision Jointly For Algebra Word Problems (2016, background): “Similar efforts have been made to develop latent structured prediction models (Yu and Joachims, 2009; Chang et al., 2013; Zettlemoyer and Collins, 2007) to ﬁnd latent semantic structures which best explain the answer given the question.”
Multitask learning for host–pathogen protein interactions (2013, methodology): “Our approach is inspired by the work from Yu and Joachims (2009) on learning structural SVMs.”
A combined pose, object, and feature model for action understanding (2012, methodology): “To solve the latent SSVM optimization problem, we use the CCCP algorithm [34], which iterates between imputing the hidden variables and learning the model weights w.”
Surrogate Benchmarks for Hyperparameter Optimization (2014, background): “The remaining prominent model types we experimented with comprised k-nearestneighbours (kNN), linear regression, least angle regression, ridge regression, SVM methods (all as implemented by scikit-learn [25]), and Bayesian neural networks (BNN) [24].”