New Machine Learning Tools for Structured Prediction - 2012

Explore new machine learning tools for structured prediction in this lecture by Veselin Stoyanov from Johns Hopkins HLTCOE. Delve into the challenges of structured prediction problems in NLP and social network analysis, focusing on Markov Random Fields (MRFs) and Probabilistic Graphical Models (PGMs). Learn about novel approaches to address approximation issues in inference, decoding, and model structure, as well as techniques for handling limited data scenarios. Discover how to minimize empirical risk in MRF-based systems using error back-propagation and local optimization, and understand the benefits of data imputation for training discriminative models with limited examples. Gain insights into training generative/discriminative hybrids that incorporate useful priors and learn from semi-supervised data. This talk is suitable for researchers and practitioners interested in advanced machine learning techniques for complex structured problems in computational linguistics and social network analysis.