Since 2018, Machine learning (ML) has disappeared from the Gartner hype cycle. It never reached plateau of productivity which IT leaders keep a close eye on before considering them adding into their capability feature set. But what does this disappearance mean ? Could it mean ML has become pervasive and businesses are benefiting by layering-on this piece of technology within their digital solution landscape? Or could it mean that ML is in the hind side delivering intelligent services , yet invisible to the end users ? The latter is perhaps true to explain the disappearance.
Machine learning today invisibly power many of our digital interactions and capabilities from Alexa to self-driving cars to cybersecurity. Its ability to learn, pre-empt and predict outcomes resembles cognitive abilities of a human mind, but not there yet ! So how does one goes about leveraging ML into their digital capability set ? Here are few approaches:
Approach 1# Realizing shortcomings from existing technology which enables certain business or technology capabilities, then exploring if ML can address these.
As an example, In the enterprise network space, techniques for classifying traffic based on port numbers on the network devices to give them specific treatment across the network has been existing for quite a while. However this technique is prone to port masquerading resulting in network compromise. This is where ML can help by classifying traffic based not only on port numbers but also by traffic payload, host behavior and flow features. Various ML techniques like supervised or unsupervised can be used to achieve this level of classification. More details can be found from Springer Open Journal
Approach 2# Explore if ML can solve problems which could not be solved before.
Encrypted traffic analysis: Its imperative now more than ever that all communications (web, voice, video) should be encrypted. However there is a risk that this encrypted traffic can contain malicious traffic specifically malware. In order to detect this certain data elements can be extracted from flow which can include sequence of packet length and time, byte distribution and ML techniques can be applied to perform behavior modeling. More details can be found from this Cisco white paper
Just like the way the human cognitive capabilities can be understood through the action of sight, speech and motion similarly ML is as well. Its in the hind side powering all the amazing digital capabilities.
I am sure there are many other vantage points which can be used to realize the applications and benefits of ML. Would love to hear from you, what is your approach to ML ?