ComputerScienceExpert

(11)

$18/per page/

About ComputerScienceExpert

Levels Tought:
Elementary,Middle School,High School,College,University,PHD

Expertise:
Applied Sciences,Calculus See all
Applied Sciences,Calculus,Chemistry,Computer Science,Environmental science,Information Systems,Science Hide all
Teaching Since: Apr 2017
Last Sign in: 5 Weeks Ago, 4 Days Ago
Questions Answered: 4870
Tutorials Posted: 4863

Education

  • MBA IT, Mater in Science and Technology
    Devry
    Jul-1996 - Jul-2000

Experience

  • Professor
    Devry University
    Mar-2010 - Oct-2016

Category > Programming Posted 08 May 2017 My Price 9.00

AI Neural Networks

Subject = AI Neural Networks.

Topic of paper = Making Intelligent antivirus.

Write 2 pages review of this paper. Use simple english , maths and AI Neural networks.

 

arXiv:1508.03096v2[cs.CR]3 Sep 2015Deep Neural Network Based Malware Detection Using Two DimensionalBinary Program FeaturesJoshua Saxe*Invincea Labs, LLCjosh.saxe@invincea.comKonstantin Berlin*Invincea Labs, LLCkberlin@invincea.comAbstractMalware remains a serious problem for corpora-tions, government agencies, and individuals, as attack-ers continue to use it as a tool to effect frequent andcostly network intrusions.Today malware detectionis still done mainly with heuristic and signature-basedmethods that struggle to keep up with malware evolu-tion. Machine learning holds the promise of automatingthe work required to detect newly discovered malwarefamilies, and could potentially learn generalizationsabout malware and benign software (benignware) thatsupport the detection of entirely new, unknown malwarefamilies. Unfortunately, few proposed machine learn-ing based malware detection methods have achieved thelow false positive rates and high scalability required todeliver deployable detectors.In this paper we introduce an approach that ad-dresses these issues, describing in reproducible detailthe deep neural network based malware detection sys-tem that Invincea has developed. Our system achievesa usable detection rate at an extremely low false posi-tive rate and scales to real world training example vol-umes on commodity hardware. Specifically, we showthat our system achieves a 95% detection rate at 0.1%false positive rate (FPR), based on more than 400,000software binaries sourced directly from our customersand internal malware databases. We achieve these re-sults by directly learning on all binaries, without anyfiltering, unpacking, or manually separating binary filesinto categories. Further, we confirm our false positiverates directly on a live stream of files coming in fromInvincea’s deployed endpoint solution, provide an esti-mate of how many new binary files we expected to seea day on an enterprise network, and describe how thatrelates to the false positive rate and translates into anintuitive threat score.Our results demonstrate that it is now feasible toquickly train and deploy a low resource, highly accurate*Authors contributed equally to the work.machine learning classification model, with false posi-tive rates that approach traditional labor intensive sig-nature based methods, while also detecting previouslyunseen malware. Since machine learning models tendto improve with larger data-sizes, we foresee deep neu-ral network classification models gaining in importanceas part of a layered network defense strategy in comingyears.1. IntroductionMalware continues to facilitate crime, espionage,and other unwanted activities on our computer net-works, as attackers use malware as a key tool their cam-paigns . One problem in computer security is thereforeto detect malware, so that it can be stopped before itcan achieve its objectives, or at least so that it can beexpunged once it has been discovered.In this vein, various categories of detection ap-proaches have been proposed, including, on the onehand, rule or signature based approaches, which requireanalysts to hand craft rules that reason over relevant datato make detections, and, on the other hand, machinelearning approaches, which automatically reason aboutmalicious and benign data to fit detection model param-eters. A middle path between these two methods is theautomatic generation of signatures. To date, the com-puter security industry, to our knowledge, has favoredmanual and automatically created rules and signaturesover machine learning and statistical methods, becauseof the low false positive rates achievable by rule andsignature-based methods.In recent years, however, a confluence of threedevelopments have increased the possibility for suc-cess in machine-learning based approaches, holding thepromise that these methods might achieve high detec-tion rates at low false positive rates without the bur-den of human signature generation required by manual

Attachments:

Answers

(11)
Status NEW Posted 08 May 2017 12:05 AM My Price 9.00

-----------

Attachments

file 1494204826-Solutions file 2.docx preview (51 words )
H-----------ell-----------o S-----------ir/-----------Mad-----------am ----------- Th-----------ank----------- yo-----------u f-----------or -----------you-----------r i-----------nte-----------res-----------t a-----------nd -----------buy-----------ing----------- my----------- po-----------ste-----------d s-----------olu-----------tio-----------n. -----------Ple-----------ase----------- pi-----------ng -----------me -----------on -----------cha-----------t I----------- am----------- on-----------lin-----------e o-----------r i-----------nbo-----------x m-----------e a----------- me-----------ssa-----------ge -----------I w-----------ill----------- be----------- qu-----------ick-----------ly -----------onl-----------ine----------- an-----------d g-----------ive----------- yo-----------u e-----------xac-----------t f-----------ile----------- an-----------d t-----------he -----------sam-----------e f-----------ile----------- is----------- al-----------so -----------sen-----------t t-----------o y-----------our----------- em-----------ail----------- th-----------at -----------is -----------reg-----------ist-----------ere-----------d o-----------n -----------THI-----------S W-----------EBS-----------ITE-----------. ----------- Th-----------ank----------- yo-----------u -----------
Not Rated(0)