Artificial Intelligence & Machine Learning , Next-Generation Technologies & Secure Development

Can Machine Learning Systems 'Overlearn'?

Sam Curry of Cybereason on When to Trust ML Systems
Sam Curry, CSO, Cybereason

Machine learning systems adapt their behavior on the basis of a feedback loop, so they can overlearn and develop blind spots, which if not understood by practitioners can lead to dangerous situations, says Sam Curry, chief security officer at Cybereason.

See Also: The Role of AI, Machine Learning in Email Security

"Don't just say you are using ML. Tell me how and where you are applying it and then how you are training it," he says. "Because that deeper understanding is what's going to let me know when to trust it and when not to."

In a video interview with Information Security Media Group at RSA Conference 2019 in San Francisco, Curry discusses:

  • The promise and dangers of AI and ML;
  • The Increase in use of automation and analytics by threat actors.

Curry, chief security officer at Cybereason, has more than 20 years of IT security industry experience. Previously, Curry served as chief technology and security officer at Arbor Networks. He also spent more than seven years at RSA, the security division of EMC, in a variety of senior management positions.


About the Author

Varun Haran

Varun Haran

Managing Director, Asia & Middle East, ISMG

Haran has been a technology journalist in the Indian market for over six years, covering the enterprise technology segment and specializing in information security. He has driven multiple industry events such as the India Computer Security Conferences (ICSC) and the first edition of the Ground Zero Summit 2013 during his stint at UBM. Prior to joining ISMG, Haran was first a reporter with TechTarget writing for SearchSecurity and SearchCIO; and later, correspondent with InformationWeek, where he covered enterprise technology-related topics for the CIO and IT practitioner.




Around the Network

Our website uses cookies. Cookies enable us to provide the best experience possible and help us understand how visitors use our website. By browsing inforisktoday.com, you agree to our use of cookies.