Defense convolutional neural network based image classification system

Artificial Intelligence (AI), such as deep learning algorithms, are widely used in modern technology and are either part of a system which uses it to accomplish tasks or operates independently to achieve certain goals. Due to the widespread usage of Artificial Intelligence, it is highly possible to...

Full description

Saved in:
Bibliographic Details
Main Author: Ng, Wing Wai
Other Authors: Chang Chip Hong
Format: Final Year Project
Language:English
Published: 2019
Subjects:
Online Access:http://hdl.handle.net/10356/78189
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:Artificial Intelligence (AI), such as deep learning algorithms, are widely used in modern technology and are either part of a system which uses it to accomplish tasks or operates independently to achieve certain goals. Due to the widespread usage of Artificial Intelligence, it is highly possible to be targeted by cyber attackers, which may force the deep learning neural network to generate undesired output, possible causing devastating consequences, such as a crash by autonomous vehicles. Hence, methods on protection of AIs are required. The project aims at developing an enhanced defensive method called Distillation, which will protect AIs from adversarial perturbation attacks. The student will be responsible for the design and training of the architecture of the AI, generate adversarial attacks and evaluate the accuracy of the AI which is protected by the Distillation method.