Skip to content

NLU 2022 Group 10 final project. We try to improve Pattern-Exploiting Training through multi-tasks training and data augmentation

Notifications You must be signed in to change notification settings

tigeryi1998/improving-pet

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

33 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Improving-PET

This is the final project of Group 10 for the course Natural Language Understanding and Computational Semantics Spring 2022.
In this project we aim to improve upon Patter-Exploiting Training (PET) by introducing data augmentation techniques as well as multi-tasks tranining

Structures

There are 3 main components: PET, AdaPET and Data Augmentation, each with their individual README file from the original authors of the paper, viewers can follow the instruction to get the example result

Pipeline

Alt text

We use EDA to generate new training examples from original dataset before feeding in to Aug-PET model

Result

Examples Method AG's Yahoo Answer
τ = 10 PET 86.7% 63.6%
τ = 10 Aug-PET 88.5% 64.4%
τ = 50 PET 86.6% 65.3%
τ = 50 Aug-PET 86.6% 65.9%
τ = 100 PET 88.1% 68.3%
τ = 100 Aug-PET 89.0% 69.1%

We present the results in ascending order of size of labelled data

About

NLU 2022 Group 10 final project. We try to improve Pattern-Exploiting Training through multi-tasks training and data augmentation

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 85.1%
  • Jupyter Notebook 13.5%
  • Shell 1.4%