ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators (ICLR 2020)

- 12 mins

Abstract

1. Introduction

figure1

2. Method

figure1

Generator

Discriminator

GAN과의 차이점

Experiments

3.1. Experimental Setup

3.2. Model Extensions

Weight sharing

Smaller Generators

figure3

Training Algorithms

3.3. Small Models

table1

3.4. Large Models

table2

table3

3.5. Efficiency Analysis

table5

figure4

5. Conclusion

Joohong Lee

Joohong Lee

Machine Learning Researcher

rss facebook twitter github youtube mail spotify instagram linkedin google pinterest medium