MACHINE
LEARNING AND
CYBERSECURITY
AUGUST 24th, 2017
Based on the work of Cristina
Vatamanu @BitDefender
SRAVAN ANKARAJU
FOUNDER & CEO
DIVERGENCE ACADEMY
Can Machine Learning Help Organizations
Improve Data Security?
YES. BUT HOW SO?
 Models based on Machine Learning are more robust.
• Machine Learning has become more proactive defense against
malware
• Most threats are file-based. Machine Learning Models are designed
for static analysis. ML integrated into protection layer of file
scanning is a proactive solution.
• In the past Signature based Threat Detection was to bypass with
some minor changes
• It gives bad guy’s headaches.
HOW SO?
File scanning protection layer is
just one layer
Are there limitations to Machine
Learning?
LIMITATION #1 – TECHNOLOGY
ITSELF
Balance of three dimensions: Detection Rate, Number of False
Positives, Performance Impact
1. NUMBER OF FALSE POSITIVES
• If you make algorithms too generic, they will be prone to False Positives
• IF you restrict them, they will cause False Negatives
2. DETECTION RATE
• Tuning becomes important
• This technology has to be backed up by other technologies such as
Whitelisting or other detection methods
3. PERFORMANCE IMPACT
• If you have to be proactive, then you have to use complex models
• Complex models will have lead for performance impact
LIMITATION #2 – TYPE OF ATTACKS
THAT A PROTECION LAYER CAN
HANDLE
• Some bypass the protection layer of file scanning
• You need models for example that scan memory pages, ones
that can intercept the vulnerabilities that are being exploited
What about Machine Learning in
Advanced Persistent Threats?
USE OF ML IN APT
• Advanced Persistent Threats are more discrete
• Guy’s in such an attack have more knowledge about their
victims
• Know what security solution is in the enterprise’s network
• They will never send a file that can be detected by security solution
• If a security solution has multiple layers of protection, ex. One
based on Dynamic Behavior, Correlating Events from Company’s
network
Is Machine Learning the CURE for all
security issues?
PURPOSE OF MACHINE LEARNING
• Machine Learning is a
Detection Tool
• Machine Learning
cannot be a protection
layer by itself
• Machine Learning can
augment the value of
protection layer
CRITERIA TO EVALUATE SECURITY
SOLUTIONS
1. What are the protection layers?
2. Are there Spam Filters, and Anti-
Phishing Filters?
3. Is there a Protection layer designed
for File Scanning or Memory Page
Scanning?
4. Are there techniques in the solution
built on Dynamic Behaviors, and
Network Anomalies?
5. Is Machine Learning being used in
any of the layers?
 All of these have to work together to
protect against different types of threats
ATTACKS ARE GETTING MORE
SOPHISTICATED
• Distributed Denial of Service (DDoS)
• Ransomware Attacks
• Insiders
• Somebody pretending to be insiders
• BYOD leads you to new challenges
• Outsource Tier-1 and Tier-2 Engagements??
• Information Assurance
DIVERGENCE ACADEMY PARTNERED WITH
DIVERGENCE ACADEMY PARTNERED WITH
DIVERGENCE ACADEMY PARTNERED WITH
DIVERGENCE ACADEMY PARTNERED WITH
SORTING A KICK OFF
SORTING B STARTS
40 HOURS A WEEK
EXTENDED A+B TOGETHER ALL
THE WAY UNTIL THE END OF A
A FINISHED, B
CONTINUES
UNTIL THE END
09/25/17 10/30/17 11/06/17 01/24/17
FUNDING OPTIONS
• Skills.fund - https://divergence.skills.fund. 36 or 60 month
loans. Living Expenses of $1500/month for three months also
available for out-of-state students.
• Workforce Innovation Opportunity Act (WIOA) funds
• Divergence Academy Tuition Installment (TADS) - 9 month
installment – 50% in the first 4 months, rest in 5 months.
JOIN TODAY TO BE THE NEXT CYBER PROFESSIONAL

Machine learning and Cybersecurity

  • 1.
    MACHINE LEARNING AND CYBERSECURITY AUGUST 24th,2017 Based on the work of Cristina Vatamanu @BitDefender SRAVAN ANKARAJU FOUNDER & CEO DIVERGENCE ACADEMY
  • 3.
    Can Machine LearningHelp Organizations Improve Data Security?
  • 4.
    YES. BUT HOWSO?  Models based on Machine Learning are more robust. • Machine Learning has become more proactive defense against malware • Most threats are file-based. Machine Learning Models are designed for static analysis. ML integrated into protection layer of file scanning is a proactive solution. • In the past Signature based Threat Detection was to bypass with some minor changes • It gives bad guy’s headaches.
  • 5.
    HOW SO? File scanningprotection layer is just one layer
  • 6.
    Are there limitationsto Machine Learning?
  • 7.
    LIMITATION #1 –TECHNOLOGY ITSELF Balance of three dimensions: Detection Rate, Number of False Positives, Performance Impact 1. NUMBER OF FALSE POSITIVES • If you make algorithms too generic, they will be prone to False Positives • IF you restrict them, they will cause False Negatives 2. DETECTION RATE • Tuning becomes important • This technology has to be backed up by other technologies such as Whitelisting or other detection methods 3. PERFORMANCE IMPACT • If you have to be proactive, then you have to use complex models • Complex models will have lead for performance impact
  • 8.
    LIMITATION #2 –TYPE OF ATTACKS THAT A PROTECION LAYER CAN HANDLE • Some bypass the protection layer of file scanning • You need models for example that scan memory pages, ones that can intercept the vulnerabilities that are being exploited
  • 9.
    What about MachineLearning in Advanced Persistent Threats?
  • 10.
    USE OF MLIN APT • Advanced Persistent Threats are more discrete • Guy’s in such an attack have more knowledge about their victims • Know what security solution is in the enterprise’s network • They will never send a file that can be detected by security solution • If a security solution has multiple layers of protection, ex. One based on Dynamic Behavior, Correlating Events from Company’s network
  • 11.
    Is Machine Learningthe CURE for all security issues?
  • 12.
    PURPOSE OF MACHINELEARNING • Machine Learning is a Detection Tool • Machine Learning cannot be a protection layer by itself • Machine Learning can augment the value of protection layer
  • 13.
    CRITERIA TO EVALUATESECURITY SOLUTIONS 1. What are the protection layers? 2. Are there Spam Filters, and Anti- Phishing Filters? 3. Is there a Protection layer designed for File Scanning or Memory Page Scanning? 4. Are there techniques in the solution built on Dynamic Behaviors, and Network Anomalies? 5. Is Machine Learning being used in any of the layers?  All of these have to work together to protect against different types of threats
  • 14.
    ATTACKS ARE GETTINGMORE SOPHISTICATED • Distributed Denial of Service (DDoS) • Ransomware Attacks • Insiders • Somebody pretending to be insiders • BYOD leads you to new challenges • Outsource Tier-1 and Tier-2 Engagements?? • Information Assurance
  • 15.
  • 16.
  • 18.
  • 19.
  • 20.
    SORTING A KICKOFF SORTING B STARTS 40 HOURS A WEEK EXTENDED A+B TOGETHER ALL THE WAY UNTIL THE END OF A A FINISHED, B CONTINUES UNTIL THE END 09/25/17 10/30/17 11/06/17 01/24/17
  • 21.
    FUNDING OPTIONS • Skills.fund- https://divergence.skills.fund. 36 or 60 month loans. Living Expenses of $1500/month for three months also available for out-of-state students. • Workforce Innovation Opportunity Act (WIOA) funds • Divergence Academy Tuition Installment (TADS) - 9 month installment – 50% in the first 4 months, rest in 5 months.
  • 22.
    JOIN TODAY TOBE THE NEXT CYBER PROFESSIONAL