Hidden Markov Support Vector Machines

By Yasemin Altun, Ioannis Tsochantaridis, Thomas Hofmann. International Conference on Machine Learning, 2003

578 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/5fe5ed2a3b50becdbbcd17e7733653d5ef6ac398

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

Easy contextual intent prediction and slot detection (2013, unknown): “els (SVM-HMMs) [17].”
Infinite Markov-Switching Maximum Entropy Discrimination Machines (2013, background): “We compare the performance of our approach to large-margin HMMs (LM) (Sha & Saul, 2007), moderate-order CRFs of 5th order (5-CRF) (Ye et al., 2009), the hidden Markov support vector machine (HMSVM) approach of Altun et al. (2004), and the iSVM approach of Zhu et al. (2011) with RBF kernels.”
Kernel-Based Learning of Hierarchical Multilabel Classification Models (2006, methodology): “Also, in the absence of regularization the max-likelihood model is likely to suffer from overfitting An alternative formulation (c.f Altun et al. 2003; Taskar et al. 2003), inspired by support vector machines, is to estimate parameters that in some sense maximize the ratio P(yi |xi ;w) P(y|xi…”
Latent hierarchical structural learning for object detection (2010, background): “The bounding box determined at position (u1, v1) of the root node and the corresponding level of the image pyramid is output as an object detection if the score F (x, ~p1) > 0 (note F (x, y) = 0 if y = −1).”
Structured Prediction, Dual Extragradient and Bregman Projections (2006, methodology): “Another line of work in learning structured prediction models aims to approximate the arising QPs via constraint generation (Altun et al., 2003; Tsochantaridis et al., 2004).”
Cutting-plane training of structural SVMs (2009, methodology): “It was shown that SVM training can be generalized to structured outputs (Altun et al. 2003; Taskar et al. 2003; Tsochantaridis et al. 2004), leading to an optimization problem that is similar to multi-class SVMs (Crammer and Singer 2001) and extending the Perceptron approach described in Collins…”
An Introduction to Conditional Random Fields (2010, background): “Maximum margin methods have been generalized to the structured case [2, 122].”
Optimal Rates for the Regularized Least-Squares Algorithm (2007, background): “This observation was our original motivation for taking into account the effective dimension: a spectral theoretical parameter which quantifies some capacity properties of ρX by means of the kernel.”
Joint segmentation and classification of human actions in video (2011, methodology): “Following [2] and inspired by HMMs, we propose to use two types of features, interactions between the observation vectors and the set of predefined states as well as the transition between states of neighboring frames along the subsegment.”
Max Margin AND/OR Graph learning for parsing the human body (2008, methodology): “They [1, 22] show that the remaining (exponentially many) constraints are guaranteed to be violated by no more than , Loop over k”
Kernel methods in machine learning (2007, unknown): “ting alternative. To that extend, we will present a general formulation of large margin methods for response variables over ﬁnite sample spaces that is based on the approach suggested by Altun et al. [6] and Taskar et al. [132]. Deﬁne r(x,y;f):=f(x,y)−max y′6= y f(x,y′)=min y′6= y log p(y|x;f) p(y′|”
Bidirectional Inference with the Easiest-First Strategy for Tagging Sequence Data (2005, background): “Max-margin methods for structured data share problems of computational cost (Altun et al., 2003).”
Maximum Margin Temporal Clustering (2012, methodology): “Constraint (9) requires segment X i ( s it ,s it +1 ] to belong to cluster y it with relatively high conﬁdence, higher than that of any other cluster by a margin.”
Large Margin Methods for Structured and Interdependent Output Variables (2005, methodology): “In contrast, we have proposed an efficient algorithm (Hofmann et al., 2002; Altun et al., 2003; Joachims, 2003) even in the case of very large output spaces, that takes advantage of the sparseness of the maximummargin solution.”
Fast Kernels for String and Tree Matching (2002, methodology): “In order to apply kernel methods one defines a measurof similarity between discrete structures via a feature mapφ : X → Hk. HereX is the set of discrete structures (eg. the set of all parse trees of a language) andHK is a Hilbert space.”
Max-Margin Markov Networks (2003, background): “By contrast, previous margin-based formulations for sequence labeling [3, 1] require an exponential number of constraints.”
Structured Sparsity in Structured Prediction (2011, background): “…involve linear combinations of features and are trained discrim-inatively; examples are conditional random ﬁelds (Lafferty et al., 2001), structured support vector machines (Altun et al., 2003; Taskar et al., 2003; Tsochantaridis et al., 2004), and the structured per-ceptron (Collins, 2002a).”
Dynamic conditional random fields: factorized probabilistic models for labeling and segmenting sequence data (2004, background): “Currently, the most popular alternative approaches to training structured discriminative models are maximum-margin training (Taskar et al., 2004a; Altun et al., 2003), and perceptron training (Collins, 2002), which has been especially popular in NLP because of its ease of implementation.”