Accession Number:

AD1069792

Title:

Draper Mining and Understanding Software Enclaves (MUSE)

Descriptive Note:

Technical Report,01 Oct 2014,31 Dec 2018

Corporate Author:

The Charles Stark Draper Laboratory, Inc Cambridge United States

Personal Author(s):

• Ellingwood,Paul
• Enxing,Hugh
• Hamilton,Lei
• Harer,Jacob
• Jost,Thomas
• Kim,Louis
• Kosta,Leo
• McConley,Marc
• Ozdemir,Onur
• Reale,Chris
• Russell,Rebecca

Report Date:

2019-03-01

Pagination or Media Count:

94.0

Abstract:

This report describes research carried out by The Charles Stark Draper Laboratory, Inc. Draper team in the Defense Advanced Research Projects Agency DARPA Mining and Understanding Software Enclaves MUSE program under contract FA8750-15-C-0242. Our focus on the MUSE program was to develop big-data analytics using machine learning for automatic vulnerability classification and repair. Key technical advancements that we contributed to the MUSE program included 1 fast and scalable machine learning-based classifiers to detect patterns in known types of software vulnerabilities 2 a generative adversarial network GAN to advance the state of the art in automated repair of common types of software vulnerabilities and 3 a data ingestion pipeline to scrape, build, and analyze millions of functions from open-source software to generate training data for learning-based algorithms.

Descriptors:

• algorithms
• big data
• artificial intelligence software
• artificial intelligence
• data mining
• feature extraction
• computer programs
• supervised machine learning
• artificial neural networks
• machine learning
• data analysis
• cybersecurity
• deep learning

Subject Categories:

• Computer Programming and Software
• Information Science

Distribution Statement:

APPROVED FOR PUBLIC RELEASE