Efficient Learning of GAN.

Date of Submission

December 2020

Date of Award

Winter 12-12-2021

Institute Name (Publisher)

Indian Statistical Institute

Document Type

Master's Dissertation

Degree Name

Master of Technology

Subject Name

Computer Science

Department

Electronics and Communication Sciences Unit (ECSU-Kolkata)

Supervisor

Pal, Nikhil Ranjan (ECSU-Kolkata; ISI)

Abstract (Summary of the Work)

GAN or Generative Adversarial Network is a combination of two deep Neural Networks in which one network acts as a generator where the other acts as a discriminator which differentiate between real and generated fake samples. There are different variants of GAN. For every variant of GAN we have to train two deep neural networks simultaneously and the hardest part about GAN is it’s training. During training many GAN models suffer various major problems like non-convergence, mode collapse, high sensitivity to the selection of hyper-parameters and vanishing gradient. In this project we tried to address the problem Mode-collapse. Where the generator generates only one or limited variants of samples irrespective of the inputs.

Comments

ProQuest Collection ID: http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&res_dat=xri:pqm&rft_dat=xri:pqdiss:28842755

Control Number

ISI-DISS-2020-06

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

DOI

http://dspace.isical.ac.in:8080/jspui/handle/10263/7154

This document is currently not available here.

Share

COinS