Learning Graphical Model Parameters with Approximate Marginal Inference

Abstract

Likelihood based-learning of graphical models faces challenges of computational-complexity and robustness to model mis-specification. This paper studies methods that fit parameters directly to maximize a measure of the accuracy of predicted marginals, taking into account both model and inference approximations at training time. Experiments on imaging problems suggest marginalization-based learning performs better than likelihood-based approximations on difficult problems where the model being fit is approximate in nature.

To Appear, IEEE Transactions on Pattern Analysis and Machine Intelligence

References in corpus (3)

• Piecewise Training for Undirected Models
• Convergent message passing algorithms - a unifying view
• Learning Convex Inference of Marginals

Cited by in corpus (35)

• Conditional Random Fields as Recurrent Neural Networks
• DeepIGeoS: A Deep Interactive Geodesic Framework for Medical Image Segmentation
• A Bayesian Filter for Multi-view 3D Multi-object Tracking with Occlusion Handling
• SparseMAP: Differentiable Sparse Structured Inference
• The structure of optimal parameters for image restoration problems
• End-to-End Learning for Structured Prediction Energy Networks
• Video Propagation Networks
• Discriminative Training of Deep Fully-connected Continuous CRF with Task-specific Loss
• Gradient Estimation with Stochastic Softmax Tricks
• Mean-Field Networks
• Structured Prediction Energy Networks
• Deep Sparse Coding Using Optimized Linear Expansion of Thresholds
• Learning Sparse High Dimensional Filters: Image Filtering, Dense CRFs and Bilateral Neural Networks
• LP-SparseMAP: Differentiable Relaxed Optimization for Sparse Structured Prediction
• Learning Linear Programs from Optimal Decisions
• A Tutorial on Modeling and Inference in Undirected Graphical Models for Hyperspectral Image Analysis
• On Image segmentation using Fractional Gradients-Learning Model Parameters using Approximate Marginal Inference
• Superpixel Convolutional Networks using Bilateral Inceptions
• Bilevel optimization in flow networks: A message-passing approach
• Real order (an)-isotropic total variation in image processing - Part II: Learning of optimal structures
• Real order (an)-isotropic total variation in image processing - Part I: analytical analysis and functional properties
• Combinatorial Optimization for Panoptic Segmentation: A Fully Differentiable Approach
• Combining Probabilistic Logic and Deep Learning for Self-Supervised Learning
• Relaxed Marginal Consistency for Differentially Private Query Answering
• Structured Learning via Logistic Regression
• Local Perturb-and-MAP for Structured Prediction
• Block Belief Propagation for Parameter Learning in Markov Random Fields
• Adversarially-learned Inference via an Ensemble of Discrete Undirected Graphical Models
• Fast and Differentiable Message Passing on Pairwise Markov Random Fields
• The Weighted Ambrosio - Tortorelli Approximation Scheme
• Soft Correspondences in Multimodal Scene Parsing
• Self-supervised self-supervision by combining deep learning and probabilistic logic
• Bethe Learning of Conditional Random Fields via MAP Decoding
• A Parametric Top-View Representation of Complex Road Scenes
• One dimensional fractional order $TGV$: Gamma-convergence and bilevel training scheme