Large Margin Methods for Structured and Interdependent Output Variables

By Ioannis Tsochantaridis, Thorsten Joachims, Thomas Hofmann, Yasemin Altun. 2005

2,337 citations recorded by the source corpus.

Subfields: Representation & Self-Supervised Learning

How researchers see this paper

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

In Representation & Self-Supervised Learning

Citations in the corpus

Quantifying Learning Guarantees for Convex but Inconsistent Surrogates (2018, background): “At the same time it is known that inconsistent surrogates often work well in practice (for example, the Crammer-Singer formulation of multi-class SVM [8], or its generalization structured SVM [20, 22]).”
On Structured Prediction Theory with Calibrated Convex Surrogate Losses (2017, methodology): “Structured SVM (SSVM) (Taskar et al., 2003; 2005; Tsochantaridis et al., 2005) is one of the most used convex surrogates for the tasks with structured output, thus, its consistency has been a question of great interest.”
Minding the Gaps for Block Frank-Wolfe Optimization of Structured SVMs (2016, methodology): “One of the most popular learning objectives for structured prediction is the structured support vector machine (Taskar et al., 2003; Tsochantaridis et al., 2005), which generalizes the classical binary SVM to problems with structured outputs.”
Block-Coordinate Frank-Wolfe Optimization for Structural SVMs (2012, methodology): “Interestingly, for structural SVMs we also show that existing batch subgradient and cutting-plane methods are special cases of Frank-Wolfe algorithms, and this leads to stronger and simpler O(1/ε) convergence rate guarantees for these existing algorithms. ar X iv :1 20 7.”
Learning Parameterized Skills (2012, background): “As previously mentioned, if important correlations between policy and task parameters are known to exist, multivariate regression models might be preferable; one possibility in such cases are Structure Support Vector Machines (Tsochantaridis et al., 2005).”
Search-based structured prediction (2009, neutral): “A 300-sentence subset of the training data set was previously used by (Tsochantaridis et al. 2005) for evaluating the SVMstruct framework in the context of sequence labeling.”
Learning Deep Structured Models (2014, methodology): “ghest scoring conﬁguration y : y = argmax y F(x;y;w): Log-linear models are a special case of this function, with F(x;y;w) = w>˚(x;y). In this setting CRFs [14], structured Support Vector Machines [27, 29] or interpolations of those [7] are typically employed to learn the linear weights w. In thi”
OptNet: Differentiable Optimization as a Layer in Neural Networks (2017, methodology): “There are two broad classes of learning approaches in structured prediction: methods that use probabilistic inference techniques (typically exploiting the fact that the gradient of the log likelihood is given by the actual feature expectations minus their expectation under the learned model (Koller ”
Large-Scale Hierarchical Text Classification with Recursively Regularized Deep Graph-CNN (2018, methodology): “We use the implementation of Hierarchical SVM (HSVM) [41] released by the authors and implement our version of Top-down Support Vector Machines (TD-SVM) [30].”
Reward Augmented Maximum Likelihood for Neural Structured Prediction (2016, neutral): “The foundations of learning structured output models were established by seminal work on graph transformer networks [21], conditional random fields (CRFs) [20], and structured large margin methods [40, 38], which demonstrate how generalization performance can be significantly improved, when one cons”
Zero-Shot Learning—A Comprehensive Evaluation of the Good, the Bad and the Ugly (2017, background): “The published code has hard-coded hyperpa-rameters operational on aPY, i.e. number of iterations, number of data points to train SVM, and one regularizer parameter γ which lead to inferior results than the ones reported here, therefore we set these parameters on validation sets.”
Zero-Shot Learning — The Good, the Bad and the Ugly (2017, background): “SJE [4] gives full weight to the top of the ranked list and is inspired from the structured SVM [36]:”
An Algorithmic Perspective on Imitation Learning (2018, methodology): “Hinge loss is a loss function often used for maximum margin optimization in classifiers such as support vector machines (SVMs) [Cortes and Vapnik, 1995].”
Search-based Neural Structured Learning for Sequential Question Answering (2017, methodology): “By leveraging a margin-based objective function, the model learning procedure resembles several structured-output learning algorithms such as structured SVMs (Tsochantaridis et al., 2005), but can take either strong or weak supervision seamlessly.”
Kernel methods in machine learning (2007, neutral): “[137], and a special instance of the above lemma is given by Joachims [78].”
Deep Learning of Representations: Looking Forward (2013, methodology): “redicted variables, we propose to investigate non-factorial posterior distributions over the observed variables, i.e., in the spirit of the recent ﬂurry of work on structured output machine learning (Tsochantaridis et al., 2005). We can think of this parametrized computational graph as a family of f”
Few-Shot Learning Through an Information Retrieval Lens (2017, methodology): “SSVM [3] is an efficient method for these tasks with the advantage of being tunable to custom task loss functions.”
Evaluation of output embeddings for fine-grained image classification (2014, methodology): “Inspired by previous work on label embedding [56] [3] [1] and structured SVMs [52] [38], we propose to use a Structured Joint Embedding (SJE) framework (Fig.”
Classical Structured Prediction Losses for Sequence to Sequence Learning (2017, background): “MaxMargin (Equation 5) is a classical margin loss for structured prediction (Taskar et al., 2003; Tsochantaridis et al., 2005) which enforces a margin between the model scores of the highest scoring candidate sequence ˆ u and a reference sequence.”