Thesis Defense: Controlling Neural Language Generation
, MIT CSAIL
Date: Tuesday, May 03, 2022
Time: 1:30 PM to 3:00 PM Note: all times are in the Eastern Time Zone
Location: 32-D463 (Star), https://mit.zoom.us/j/97034696195
Event Type: Thesis Defense
Host: Regina Barzilay, MIT CSAIL
Contact: Tianxiao Shen, firstname.lastname@example.org
Speaker URL: http://people.csail.mit.edu/tianxiao/
TALK: Thesis Defense: Controlling Neural Language Generation
Thesis advisors: Tommi Jaakkola, Regina Barzilay
Thesis committee: Yoon Kim
Language models have achieved success in a wide range of applications such as machine translation, dialogue generation, and writing autocompletion. However, typical models operate in a left-to-right, unconstrained fashion with limited control over what is generated. In this talk, I’ll discuss three directions to control language generation. First, we’ll leverage distributional alignment to perform style transfer from non-parallel text. Next, we’ll study confounding factors and use invariance to isolate them to transfer text in the desired direction. Lastly, we’ll look at flexible generation location control, for which we developed sequence models that dynamically create and fill in blanks during the generation process.
AI & Machine Learning
Created by Tianxiao Shen at Wednesday, April 27, 2022 at 2:21 PM.