Homework12_Solution (1)

.pdf

School

Rutgers University *

*We aren’t endorsed by this school

Course

446

Subject

Computer Science

Date

Jan 9, 2024

Type

pdf

Pages

1

Uploaded by dan47433

Report
ECE 446/579:04 (Spring 2023) Homework #12 Solution 1. (Deep Learning System Security - Attacks) In the lecture we discussed two papers related to attacks on deep learning accelerators: 1) Y. Zhao et al., Memory Trojan Attack on Neural Network Accelerators, DATE 2019. 2) W. Hua et al., Reverse Engineering Convolutional Neural Networks Through Side-channel Information Leaks, DAC 2018 Please comment on the similarities and differences between the two papers in terms of the properties of the security consequences (e.g., breaking confidentiality or integrity) and the attack methods (e.g., physical attack, side channel attack, or a mix of multiple methods). 1) Memory Trojan: hardware approach, breaking confidentiality/integrity, involving side channel attack (memory access pattern); 2) Reverse engineering: software approach, breaking confidentiality, involving side channel attack (memory access pattern). 2. (Deep Learning System Security – Defense Techniques) Can you please describe at least one countermeasure for the software attack (i.e., adversarial attack), memory Trojan attack, and reverse engineering attack on deep learning systems? 1) Countermeasures for adversarial attack: adversarial training, perturbation removal, etc. 2) Countermeasures for memory Trojan or reverse engineering attacks that rely on memory access patterns: hiding the memory access patterns, e.g., oblivious RAM (ORAM).
Discover more documents: Sign up today!
Unlock a world of knowledge! Explore tailored content for a richer learning experience. Here's what you'll get:
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help