Presentation: Smoothing Methods with Ajay Bapat

Abstract:

When modeling language with probabilities an unseen word (i.e. a zero probability) will transform a perfectly good sequence of numbers into a useless one. Smoothing methods present on way to avoid this problem.

In this talk I'll be giving an overview of smoothing methods, with some theory and motivation behind different smoothing methods. The topics I plan on covering are listed below:

• Additive smoothing.

• Estimating probabilities from counts using held-out set and Good-Turing method.

• Jelinek-Mercer smoothing

• Katz Back-off model

• Kneser-Ney model

Presenter Bio:

Ajay Bapat works in the search and data team at a startup in Boulder.