Tractable Learning and Inference with High-Order Representations
Aron Culotta, Andrew McCallum · ScholarWorks@UMassAmherst (University of Massachusetts Amherst) · 2006
Representing high-order interactions in data often results in large models with an in-tractable number of hidden variables. In these models, inference and learning must op-erate without instantiating the entire set of variables. This paper presents a Metropolis-Hastings sampling approach to address this issue, and proposes new methods to discrimi-natively estimate the proposal and target dis-tribution of the sampler using a ranking func-tion over configurations. We demonstrate our approach on the task of paper and author deduplication, showing that our method en-ables complex, advantageous representations of the data while maintaining tractable learn-ing and inference procedures. 1.