Hoifung Poon

Address:
Dept. of Computer Science & Engineering
University of Washington
Box 352350
Seattle, WA 98195-2350

Telephone: (206) 543-8086
Fax: (206) 543-2969
Email: <first name > at cs dot washington dot edu
Office: 318 Allen Center

I am now a researcher in the NLP group at Microsoft Research. Please visit my new website.

In June, I defended my Ph.D. dissertation in the Department of Computer Science and Engineering at the University of Washington. My advisor is Pedro Domingos.

My research interest are in statistical relational learning (SRL) and natural language processing (NLP). SRL is an emerging direction that combines logic and probability to handle both complexity and uncertainty in large-scale machine learning. Recent work has shown that SRL can compactly specify very large and complex probabilistic models and substantially improve predictive accuracy. In addition, my thesis work shows that SRL can help breach the bottleneck of large-scale machine learning: the scarcity of labeled examples. To compensate for the lack of label examples, my research has applied SRL to leverage indirect supervision via joint inference, where the labels of some objects can be used to predict the labels of others. This has enabled me to develop state-of-the-art unsupervised learning algorithms for a variety of NLP problems ranging from segmenting words to inducing ontologies.

These advances are based on Markov logic, a leading unifying framework for SRL. I have developed efficient learning and inference algorithms for Markov logic that can handle problems with millions of variables and billions of dependencies.

I am particularly excited about the long-standing goal of harnessing human knowledge by automatically understanding texts, a.k.a. machine reading. I have developed USP, an end-to-end machine reading system that can read text, extract knowledge, and answer questions, all without any labeled examples. In a machine reading experiment, USP extracted five times as many correct answers compared to state of the art such as TextRunner, and raised accuracy from below 60% to 91%.

I recently gave a tutorial in NAACL-2010 on Markov logic in natural language processing. It was the most popular tutorial in the conference (thanks all who came!). Here are the slides.

Publications

Sum-Product Networks: A New Deep Architecture. [Paper] [Slides] [Download code and results]
Hoifung Poon and Pedro Domingos.
In Proceedings of the Twenty-Seventh Conference on Uncertainty in Artificial Intelligence, (UAI), 2011.
Acceptance rate: 34%. (Plenary acceptance: 9%)
Best Paper Award

Unsupervised Ontology Induction from Text. [Paper] [Slides]
Hoifung Poon and Pedro Domingos.
In Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL), 2010.
Acceptance rate: 25%.

Joint Inference for Knowledge Extraction from Biomedical Literature. [Paper] [Slides]
Hoifung Poon and Lucy Vanderwende.
In Proceedings of the North American Chapter of the Association for Computational Linguistics - Human Language Technologies Conference (NAACL-HLT), 2010.
Acceptance rate: 31%.

Unsupervised Semantic Parsing. [Paper] [Slides] [Download data and code]
Hoifung Poon and Pedro Domingos.
In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), 2009.
Acceptance rate: 20%.
Best Paper Award

Unsupervised Morphological Segmentation with Log-Linear Models. [Paper] [Slides]
Hoifung Poon, Colin Cherry, and Kristina Toutanova.
In Proceedings of the North American Chapter of the Association for Computational Linguistics - Human Language Technologies Conference (NAACL-HLT), 2009.
Acceptance rate: 29%.
Best Paper Award & IBM Best Student Paper Award

Language ID in the Context of Harvesting Language Data off the Web. [Paper]
Fei Xia, William Lewis and Hoifung Poon.
In Proceedings of the Conference of European Association for Computational Linguistics (EACL), 2009.
Acceptance rate: 28%.

Joint Unsupervised Coreference Resolution with Markov Logic. [Paper]
Hoifung Poon and Pedro Domingos.
In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), 2008.
Acceptance rate: 20%.

A General Method for Reducing the Complexity of Relational Inference and its Application to MCMC. [Paper]
Hoifung Poon, Pedro Domingos, and Marc Sumner.
In Proceedings of the Twenty-Third AAAI Conference on Artificial Intelligence (AAAI), 2008.
Acceptance rate: 24%.

Markov Logic. [Book Chapter]
Pedro Domingos, Stanley Kok, Daniel Lowd, Hoifung Poon, Matthew Richardson, Parag Singla.
In L. De Raedt, P. Frasconi, K. Kersting and S. Muggleton (eds.), Probabilistic Inductive Logic Programming, 2008.

Joint Inference in Information Extraction. [Paper] [Online Appendix]
Hoifung Poon and Pedro Domingos.
In Proceedings of the Twenty-Second National Conference on Artificial Intelligence (AAAI), 2007.
Acceptance rate: 27%.

Sound and Efficient Inference with Probabilistic and Deterministic Dependencies. [Paper] [Slides]
Hoifung Poon and Pedro Domingos.
In Proceedings of the Twenty-First National Conference on Artificial Intelligence (AAAI), 2006.
Acceptance rate: 21%.

Unifying Logical and Statistical AI. [Paper]
Pedro Domingos, Stanley Kok, Hoifung Poon, Matthew Richardson, Parag Singla.
In Proceedings of the Twenty-First National Conference on Artificial Intelligence (AAAI), 2006.
Invited paper.

Software

Sum-Product Networks

Unsupervised Semantic Parsing

The Alchemy System for Statistical Relational AI

Last modified: September, 2011