Learning in SMT | The usual presentation of MIRA’s optimization problem is given as a quadratic program: |
Learning in SMT | While solving the optimization problem relies on computing the margin between the correct output yi, and y’, in SMT our decoder is often incapable of producing the reference translation, i.e. |
Learning in SMT | In this setting, the optimization problem becomes: |
The Relative Margin Machine in SMT | The online latent structured soft relative margin optimization problem is then: |
The Relative Margin Machine in SMT | The dual in Equation (5) can be optimized using a cutting plane algorithm, an effective method for solving a relaxed optimization problem in the dual, used in Structured SVM, MIRA, and RMM (Tsochantaridis et al., 2004; Chiang, 2012; Shivaswamy and Jebara, 2009a). |
Our Approach | By solving the optimization problem in equation (4), we can get the reduced representation of terms and questions. |
Our Approach | ,U p fixed, the update of Up amounts to the following optimization problem: |
Our Approach | Thus, the optimization of equation (5) can be decomposed into Mp optimization problems that can be solved independently, with each corresponding to one row of Up: |
Compressive Summarization | In previous work, the optimization problem in Eq. |
Extractive Summarization | By designing a quality score function g : {0, 1}N —> R, this can be cast as a global optimization problem with a knapsack constraint: |
Extractive Summarization | 1, one obtains the following Boolean optimization problem: |
Introduction | els (maximizing relevance, and penalizing redundancy) lead to submodular optimization problems (Lin and Bilmes, 2010), which are NP-hard but ap-proximable through greedy algorithms; learning is possible with standard structured prediction algorithms (Sipos et al., 2012; Lin and Bilmes, 2012). |
Introduction | Technically, instead of doing standard Bayesian inference via Bayes’ rule, which requires a normalized likelihood model, we propose to do regularized Bayesian inference (Zhu et al., 2011; Zhu et al., 2013b) via solving an optimization problem , where the posterior regularization is defined as an expectation of a logistic loss, a surrogate loss of the expected misclassification error; and a regularization parameter is introduced to balance the surrogate classification loss (i.e., the response log-likelihood) and the word likelihood. |
Introduction | 2 introduces logistic supervised topic models as a general optimization problem . |
Logistic Supervised Topic Models | As noticed in (Jiang et al., 2012), the posterior distribution by Bayes’ rule is equivalent to the solution of an information theoretical optimization problem |
Logistic Supervised Topic Models | supervised topic model (MedLDA) (Jiang et al., 2012), which has the same form of the optimization problems . |
Budgeted Submodular Maximization with Cost Function | We argue that our optimization problem can be regarded as an extraction of subtrees rooted at a given node from a directed graph, instead of from a tree. |
Conclusions and Future Work | We formalized a query-oriented summarization, which is a task in which one simultaneously performs sentence compression and extraction, as a new optimization problem : budgeted monotone nondecreasing submodular function maximization with a cost function. |
Joint Model of Extraction and Compression | An optimization problem with this objective function cannot be regarded as an ILP problem because it contains nonlinear terms. |