Abstract:Real-world problems may contain latent dependencies (ie, hidden sub-structures) that are difficult to capture with conventional structured classifiers, such as conditional random fields. In such cases, models that exploit latent variables are advantageous in learning, and conditional latent variable models have been applied successfully into real-world tasks in natural language processing and vision processing communities, including syntactic parsing and vision recognition. In the first part of this thesis, I perform experiments in a variety of tasks to confirm the advantages of conditional latent variable models, and investigate what kind of latent dependencies are learned by the model. While the experiments confirmed the advantages of conditional latent variable models, the same experiments revealed two problems in applying such models for practical usages. First, establishing an efficient inference method on latent conditional models remains an open question. Second, because of the incorporation of latent variables, training a latent conditional model brings a heavy computational cost. To deal with those critical problems, I propose efficient methods for inference and training on latent conditional models.In the inference stage, I propose the “latent-dynamic inference”, which is able to produce the exact optimal label sequence on latent conditional models by systematically combining efficient search strategy (the A* search) and dynamic programming (the Forward-Backward method). I also describe a straightforward solution on approximating the exact method, and show that the approximated version performs as well as the exact one, while the speed …

EFFICIENT INFERENCE AND TRAINING FOR CONDITIONAL LATENT VARIABLE MODELS¡¢£¤¥¦ § ©

Sequential Labeling with Latent Variables: an Exact Inference Algorithm and Its Efficient Approximation

Efficient Probabilistic Latent Semantic Analysis with Sparsity Control

Exact Decoding on Latent Variable Conditional Models is NP-Hard

Robust Mixture Bayesian Latent Variable Regression with Structural Sparsity and Application to Inferential Sensing of Quality Variables

Analyzing and Improving Supervised Nonlinear Dynamical Probabilistic Latent Variable Model for Inferential Sensors

Semi-supervised Mixture of Latent Factor Analysis Models with Application to Online Key Variable Estimation

Learning Global Controller in Latent Space for Parameter-Efficient Fine-Tuning

Fast Gradient-Based Inference with Continuous Latent Variable Models in Auxiliary Form

Modeling Latent-Dynamic in Shallow Parsing

Latent Theme Dictionary Model for Finding Co-occurrent Patterns in Process Data

Graphical model selection with latent variables

Discriminative conditional restricted Boltzmann machine for discrete choice and latent variable modelling

A Unified Theory of Exact Inference and Learning in Exponential Family Latent Variable Models

Nonsparse learning with latent variables

Joint Learning Of Semantic And Latent Attributes

Conditional topical coding: an efficient topic model conditioned on rich features.

Latent state recognition by an enhanced hidden Markov model

A fast self-adaptive intuitionistic fuzzy latent factor model

Transfer Learning for Latent Variable Network Models

Latent Dirichlet classification:A new method for object detection