We consider the problem of parameter estimation using weakly supervised datasets, where a training sample consists of the input and a partially specified annotation, which we refer to as the output. The missing information in the annotation is modeled using latent variables. Previous methods overburden a single distribution with two separate tasks: (i) modeling the uncertainty in the latent var...

The basic idea of factor analysis is the following. For a given set of manifest variables x1, . . . , xp one wants to find a set of latent variables ξ1, . . . , ξk, fewer in number than the manifest variables, that contain essentially the same information. The latent variables are supposed to account for the dependencies among the manifest variables in the sense that if the latent variables are...

. Grammars with Latent Variables Given a treebank over a set of categories learn an optimally refined grammar for parsing. The parameters can be learned with an Expectation Maximization algorithm. The E-Step involves computing expectations over derivations corresponding to the observed trees. These expectations are normalized in the M-Step to update the rewrite probabilities: Grammar Learning T...

Plausible values (PVs) are a standard multiple imputation tool for analysis of large education survey data, which measures latent proficiency variables. When latent proficiency is the dependent variable, we reconsider the standard institutionally generated PV methodology and find it applies with greater generality than shown previously. When latent proficiency is an independent variable, we sho...

Previous studies showed that liberals and conservatives differ in cognitive style. Liberals are more flexible, and tolerant of complexity and novelty, whereas conservatives are more rigid, are more resistant to change, and prefer clear answers. We administered a set of compound remote associate problems, a task extensively used to differentiate problem-solving styles (via insight or analysis). ...

Recently there has been substantial interest in using spectral methods to learn generative sequence models like HMMs. Spectral methods are attractive as they provide globally consistent estimates of the model parameters and are very fast and scalable, unlike EM methods, which can get stuck in local minima. In this paper, we present a novel extension of this class of spectral methods to learn de...

The contribution of generalized method of moments (Hansen and Singleton, 1982) was to allow frequentist inference regarding the parameters of a nonlinear structural model without having to solve the model. Provided there were no latent variables. The contribution of this paper is the same. With latent variables.

We review our recent work, which includes several disparate approaches and tasks, all with the goal of using latent articulatory structure while learning discriminatively.

We derive constructive locally efficient estimators in semiparametric measurement error models. The setting is one where the likelihood function depends on variables measured with and without error, where the variables measured without error can be modelled nonparametrically. The algorithm is based on backfitting. We show that if one adopts a parametric model for the latent variable measured wi...