Deep learning, artificial neural networks, deep neural networks and artificial intelligence; these are the buzzwords in vogue at the moment. This is mainly due to the amazing results exhibited by these technologies across many applications: computer vision (including object recognition, face recognition, etc), speech and speaker recognition, natural language understanding and processing and even more advanced combinations of these results (e.g., coupling deep learning based object recognition with deep learning based language models for automatic image annotation). In this deep learning revolution, Deep Instinct is the first company to apply deep learning to cybersecurity. Deep Instinct offers proactive defense that protects against known and unknown malware in real-time, across an organization’s mobile devices, desktops, and servers.
Neural networks have been around since the 1960’s but their popularity wore off in late 60’s and 70’s due to lack of training methods for neural networks beyond the very simple ones, which rendered them impractical for many ‘real world’ applications. In the 1980’s, several mathematical contributions led to a resurgence of neural networks. While neural networks achieved good results in the 1990s and early 2000s, they were largely overshadowed by methods such as support vector machines (SVM). While it was already known back then that training deep neural networks with many layers would provide substantially superior results, training such networks was not feasible due to both algorithmic and computational limitations. However, during the past few years, neural networks, in the form of deep learning, have finally made a huge comeback, due to both major algorithmic improvements, and their implementation on graphical processing units (GPUs) which provide tremendously improved computational capabilities Deep learning, being the first family of methods within machine learning which is truly domain agnostic and does not require feature engineering, easily lends itself to successful application in nearly every field.
Due to the high barrier of entry for creating deep learning infrastructures and employing them, currently there are few companies using state-of-the-art deep learning capabilities. The most notable examples are Google, Facebook, Baidu, and Microsoft, which use deep learning for obtaining groundbreaking improvements in nearly every field that involves images, audio, or text as the raw data.
Deep Instinct is the first company to apply deep learning to cybersecurity. We have been training our models to instantly differentiate between legitimate and malicious code, from any operating system (Android, Windows, iOS) on any platform (desktop, servers and mobile devices). Similar to every field in which deep learning is applied, when testing the detection rates of zero-day malware of our solution with 61 leading solutions on the market, the results are outstanding: in all the benchmarks conducted, Deep Instinct achieved a 98+% detection rate, while the best detection rate out of those competing solutions tested against was 79%. This 20+% gap observed here is consistent with the improvements obtained by deep learning in other domains (e.g., classification error rate for the challenging ImageNet dataset for computer vision dropped from 25% by the best classical computer vision methods to 4% obtained by deep learning, better than humans which achieve 5% error rate).
By applying deep learning to cybersecurity for the first time, Deep Instinct provides protection that learns on its own to differentiate between malicious and legitimate malware, enabling immediate prediction of malware that has never been seen before. Since the learning process is data agnostic, zero-day and APT attacks are immediately detected and prevented across the enterprise’s endpoints, servers, and mobile devices.