By Morgan Jay, Area Vice President at Imperva.
We often question what drives the success behind enormous companies like Google and Amazon. A large part of the answer is machine learning. These companies have quickly adopted machine learning, finding smarter ways to apply it and changing the dynamic of how they work. With the extra analytical muscle that machine learning affords, they’re able to drive more intelligent and innovative projects which – let’s be honest – just work.
The result of the dominance of these companies is that we have become more familiar with the capabilities of machine learning than ever before. With mobile phones knowing us better than we know ourselves, and enterprise technologies predicting every next step, machine learning is clearly going to be a key part of our future.
It should come as no surprise, then, that the potential of adopting machine learning in the cybersecurity sector is now being recognised. As organisations collect increasingly more data, they are also met with a corresponding growth in security threats that they need to cope with. Therefore, developers are turning to alternative, smarter and more efficient ways to protect sensitive business data. So how can machine learning be applied to cybersecurity where it offers the most value?
The ideal use cases for machine learning are those that involve large data sets that would have been too time consuming to analyse in the past. These systems adapt and grow from experience, in a similar way to how humans hone their skills over time. Also like humans, machine learning will also be incorrect to a certain percentage, so they can’t completely replace human beings for decisions that require 100% certainty.
Crucially, machine learning applications require large amounts of data as fuel to learn from. This is why cybersecurity is such fertile ground because the datasets that cybersecurity systems are generating can be gargantuan at times. When we also consider that the cybersecurity field is facing a shortfall of 1.8 million qualified professionals by 2022, the entire sector is under pressure to find new solutions.
Because machine learning applications can learn new skills much faster than humans, they can close many of the skills gaps we are likely to face. In a security setting, machine learning enables us to detect patterns and establish baseline data access behaviour using algorithms that learn through training or observation. This is particularly useful when we consider the ever-present threat posed to critical data by careless or malicious insiders.
Machine learning enables security teams to efficiently determine whether each access behaviour is normal, and decide whether it should still be allowed to continue. The sheer volume of data involved within data access logs would have made these decisions impossible previously, but machine learning can quickly process this data to provides us with clear and contextual results.
By establishing and a baseline of data access patterns across your organisation, machine learning uses pattern recognition to identify normal behaviour for individuals in specific groups. Once an organisation’s normal data access patterns are identified, it becomes a far simpler task to filter those careless or malicious behaviours that threaten to compromise enterprise data.
Machine learning can also help us to climb out from under the avalanche of alerts that regularly bury our security teams. Imperva surveyed 179 IT professionals, and 29% of them told us they receive more than a million threat alerts each day, and more than half of respondents (55%) told us that they are dealing with more than 10,000 alerts each day.
Coping with so many potential threats on a daily basis can quickly cause alert fatigue. They receive so many alerts, they simply cannot investigate them all, and have little idea on how to prioritise them. Machine learning helps these teams to categorise alerts, so they know the high-risk alerts to start with, in order to maximise the human effort required to investigate and mitigate.
Reducing alert fatigue and efficiently monitoring user access are just some of the ways machine learning can transform the cybersecurity landscape. We are just beginning to see the potential for machine learning, and more is yet to come. As newer and more effective ways of adopting these new technologies are uncovered, the future of cybersecurity is looking bright. While we cannot rely on these technologies for the talent shortfall the industry is facing, our most immediate and challenging problems can be solved if we make the most of this versatile technology.
