As with cloud-washing, I’ve become weary of ‘AI/ML-washing’. I’d go further than this article and suggest that not every situation is appropriate for the application of deep learning techniques. Bear that in mind the next time a vendor adds ‘now with added artifcial intelligence/machine learning/deep learning‘ to their marketing blurb…:
Deep learning is not an entirely new concept. Tech giants like Google, Facebook and Microsoft already have their own frameworks, including TensorFlow, PyTorch and CNTK.
These publicly available frameworks have significantly boosted deep learning research as developers build applications and programmes from them without having to write low-level code.
While deep learning frameworks such as the above are being applied across many sectors, there are limitations when it comes to cybersecurity. These established frameworks cannot serve as an effective neural network for learning and predicting cyberattacks, and are limited in the three following ways:
Inefficient predictions: Publicly available frameworks are unable to deliver on the special performance requirements for the predictions that cybersecurity demands. This is because generic frameworks usually require dedicated hardware to enable real-time predictions. However, when deep learning is used in the real world, performance can be limited when running on devices with standard CPUs without dedicated supporting hardware. This severely hampers the ability to deliver accurate and timely predictions of cyberattacks.
Not production-ready: Features on the publicly available frameworks are mostly research rather than production oriented, and it is these features that compromise performance. For example, the features that may enhance translation programmes, providing reasonable performance at speeds of 0.1 or 0.2 seconds, result in a performance penalty in terms of speed and memory when running on thousands of open files at any given point, as well as their dependence on dozens of external libraries.
These applications also require a large memory footprint. While not a big concern for research-driven activities, it can be prohibitive in production environments, presenting heavier workloads on servers and low-power laptops. An additional implication of the research orientation of these frameworks is that the algorithms used in the cybersecurity field are not yet implemented in the publicly available frameworks.
Not sufficiently secure: Finally, the many features in generic deep learning frameworks offer a large attack surface, making these frameworks more vulnerable to the exact cyberattacks that developers work to prevent.
Solution: a specific neural network designed for cybersecurity
To overcome these limitations, enterprises need to integrate specific deep learning frameworks designed for cybersecurity. The computing infrastructure and algorithms within these frameworks are optimized for detecting and protecting businesses from any known and unknown malware in real-time. The highly comprehensive predictive capabilities of specific deep learning frameworks lead to:
- Highest accuracy with real-time predictions
- Optimized for inference mode
- Designed for commercial production
- Highest levels of security
Delivering unmatched accuracy and efficacy, specific deep learning frameworks offer enterprises cybersecurity solutions that provide end-point and mobile prevention, and detection-and-response against any file-based or file-less attack, across any device and operating system, on one unified platform.