Lifelong Twin Generative Adversarial Networks

Fei Ye; Adrian Gheorghe Bors

Lifelong Twin Generative Adversarial Networks

Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

In this paper, we propose a new continuously learning generative
model, called the Lifelong Twin Generative Adversarial
Networks (LT-GANs). LT-GANs learns a sequence of tasks
from several databases and its architecture consists of three
components: two identical generators, namely the Teacher
and Assistant, and one Discriminator. In order to allow for
the LT-GANs to learn new concepts without forgetting, we
introduce a new lifelong training approach, namely Lifelong
Adversarial Knowledge Distillation (LAKD), which encourages
the Teacher and Assistant to alternately teach each other,
while learning a new database. This training approach favours
transferring knowledge from a more knowledgeable player to
another player which knows less information about a previously
given task.

Original language	English
Title of host publication	Proc. of IEEE International Conference on Image Processing (ICIP)
Publisher	IEEE
Number of pages	5
Publication status	Published - 20 Sept 2021

Access to Document

ICIP2021_Lifelong_Twin_GANsAccepted author manuscript, 601 KB

Cite this

@inproceedings{de73fb94575440ee8fcb314a1b54591f,

title = "Lifelong Twin Generative Adversarial Networks",

abstract = "In this paper, we propose a new continuously learning generativemodel, called the Lifelong Twin Generative AdversarialNetworks (LT-GANs). LT-GANs learns a sequence of tasksfrom several databases and its architecture consists of threecomponents: two identical generators, namely the Teacherand Assistant, and one Discriminator. In order to allow forthe LT-GANs to learn new concepts without forgetting, weintroduce a new lifelong training approach, namely LifelongAdversarial Knowledge Distillation (LAKD), which encouragesthe Teacher and Assistant to alternately teach each other,while learning a new database. This training approach favourstransferring knowledge from a more knowledgeable player toanother player which knows less information about a previouslygiven task.",

author = "Fei Ye and Bors, {Adrian Gheorghe}",

year = "2021",

month = sep,

day = "20",

language = "English",

booktitle = "Proc. of IEEE International Conference on Image Processing (ICIP)",

publisher = "IEEE",

address = "United States",

}

TY - GEN

T1 - Lifelong Twin Generative Adversarial Networks

AU - Ye, Fei

AU - Bors, Adrian Gheorghe

PY - 2021/9/20

Y1 - 2021/9/20

N2 - In this paper, we propose a new continuously learning generativemodel, called the Lifelong Twin Generative AdversarialNetworks (LT-GANs). LT-GANs learns a sequence of tasksfrom several databases and its architecture consists of threecomponents: two identical generators, namely the Teacherand Assistant, and one Discriminator. In order to allow forthe LT-GANs to learn new concepts without forgetting, weintroduce a new lifelong training approach, namely LifelongAdversarial Knowledge Distillation (LAKD), which encouragesthe Teacher and Assistant to alternately teach each other,while learning a new database. This training approach favourstransferring knowledge from a more knowledgeable player toanother player which knows less information about a previouslygiven task.

AB - In this paper, we propose a new continuously learning generativemodel, called the Lifelong Twin Generative AdversarialNetworks (LT-GANs). LT-GANs learns a sequence of tasksfrom several databases and its architecture consists of threecomponents: two identical generators, namely the Teacherand Assistant, and one Discriminator. In order to allow forthe LT-GANs to learn new concepts without forgetting, weintroduce a new lifelong training approach, namely LifelongAdversarial Knowledge Distillation (LAKD), which encouragesthe Teacher and Assistant to alternately teach each other,while learning a new database. This training approach favourstransferring knowledge from a more knowledgeable player toanother player which knows less information about a previouslygiven task.

M3 - Conference contribution

BT - Proc. of IEEE International Conference on Image Processing (ICIP)

PB - IEEE

ER -