Docket #: S04-319
Fast feature selection method and system for maximum entropy modeling
Maximum Entropy (ME) modeling is a general statistical modeling paradigm that may be applied in language modeling and natural language processing to predict linguistic behavior by incorporating various informative features, each encoding some linguistically statistical event, from a corpus of data into a common framework of conditional models.
The present invention is intended to provide a fast method for selecting high quality features for Maximum Entropy (ME) modeling that may be applied in areas of statistical modeling and linear regression, ranging from language understanding and bio-informatics to stock market prediction. In this regard, the fast feature selection method of the present invention may build compact, high-quality, robust models and make feasible many previously impractical tasks.
Applications
- Statistical modeling and linear regression
- Language understanding
- Bio-informatics
- Stock market prediction
Advantages
- Fast methodology
- Robust
Publications
- Zhou et al. ACL Anthology (2003) "Fast algorithm for feature selection in conditional maximum entropy modeling"
Patents
- Published Application: WO2005008365
- Published Application: 20050021317
- Issued: 7,324,927 (USA)
Similar Technologies
-
Phrasal: A Statistical Phrase-based Machine Translation System S15-036Phrasal: A Statistical Phrase-based Machine Translation System
-
Trainable Analog Hardware Platform for Recurrent Neural Networks (RNN) S19-165Trainable Analog Hardware Platform for Recurrent Neural Networks (RNN)
-
State-of-the-Art Graph Diffusion Transformer for Natural Language Processing S20-271State-of-the-Art Graph Diffusion Transformer for Natural Language Processing