Deep Learning Holds Promise to a Safer Cybersecurity Future, Says Report From MIT and Deep Instinct
If the world is ever to get a decisive leg up on cybersecurity, it’s mindset must change from one of detection to that of prevention. That may be easier said than done, but a new report from the Massachusetts Institute of Technology (MIT), produced in cooperation with Deep Instinct, holds the promise of a safer cyber future… a new line of defense called “deep learning.”
Deep Learning: Prevention Over Detection
More common cybersecurity technologies, such as endpoint detection and response (EDR), simply identify, track, record and contain a cyber threat once it has already entered an environment. While machine learning cybersecurity solutions are a critical component of a security strategy, like EDR, it cannot proactively defend against sophisticated attacks without constant human tweaking, according to the MIT Technology Review Insights report.
However, deep learning can mimic the functionality and connectivity of neurons in the human brain. This capability enables neural networks to independently learn from raw and un-curated data and automatically recognize unknown threats, the report explains. As such, deep learning is a powerful solution that can accurately identify highly sophisticated attack patterns at record speeds.
“Deep learning is the only family of algorithms that works on raw data to identify cybersecurity threats with unmatched speed and accuracy,” Deep Instinct CEO Guy Caspi says in the report.
The report explains that “deep learning addresses the limitations of machine learning by circumventing the need for highly skilled and experienced data scientists to manually feed a solution data sets.” Thus, deep learning reduces costs and presents significant business benefits.
Deep Learning Trains Itself
A deep learning model — one specifically developed for cybersecurity — can absorb and process vast volumes of raw data to fully train the system, the report explains. Once trained, these neural networks become autonomous and do not require constant human intervention.
A combination of raw data, based learning methodology and larger data sets, means that deep learning can evolve to accurately identify much more complex patterns than machine learning — and considerably faster.
Deep learning’s ability to predict the threat of “adversarial AI” is yet another advantage, the report says. “Adversarial machine learning” is a technique that tricks AI models by feeding them deceptive data to intentionally exploit the way traditional machine learning–based solutions work. Adversarial AI can spot a bias that will bypass detection capabilities and deceive the user into accepting malicious files as benign.
A deep learning model can be deployed on any endpoint using negligible processing resources, the report says. Thus, deep learning enables a substantially higher accuracy rate in identifying malware and other fileless threats and at a much lower false positive rate.
The report notes that because deep learning is agnostic to file types, it can be applied to any file format or operating system, without requiring substantial modifications or adaptations.