Neural Classification of Malware-As-Video with Considerations for In-Hardware Inferencing

Abstract

The objective of this thesis is to explore the classification of assembly code as benign or malicious through the use of neural networks, and while building these networks, giving consideration to the creation of malware detecting hardware. Neural networks have become a go-to solution in many fields due to their ability to learn from an enormous number of features. Fully entrusting security to a neural network may be unwise due to issues with bias in training data and the ultimately unknowable nature of how the network makes a classification. If a proficient system is achieved for low cost in terms of memory or time, however, it could be another tool in the toolbox for fighting malware.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jul 01, 2019
Accession Number
AD1076695

Entities

People

  • Michael Santacroce

Organizations

  • University of Cincinnati

Tags

Communities of Interest

  • Cyber

DTIC Thesaurus Topics

  • Air Force
  • Artificial Intelligence Software
  • Automata Theory
  • Computer Languages
  • Computer Science
  • Computers
  • Convolutional Neural Networks
  • Dimensionality Reduction
  • Electrical Engineering
  • Information Science
  • Machine Learning
  • Network Science
  • Neural Networks
  • Operating Systems
  • Recurrent Neural Networks
  • Supervised Machine Learning
  • Training

Fields of Study

  • Computer science

Readers

  • Cybersecurity.
  • Neural Network Machine Learning.
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks
  • Cyber