Just about every day there is a press release where “big data,” “real-time” and “analytics” - along with machine learning - are being advertised as a breakthrough technologies that will enable business and operational transformation for mobile service providers. While it is true that mobile data analytics create new opportunities, with the notable exception of the nascent introduction of machine learning to the “analytics toolkit,” the technology is far from revolutionary.
While several recent breakthroughs have made the application of machine learning to the more general field of “predictive analytics,” standard data analytics rely on the same tools and technologies that have driven our computer programming models for the better part of the last century. With the proper data and an understanding of the output, or result, we can create a program that replicates that output (within some probabilistic bounds). Analytical programming is the same – we can create an algorithm that looks at patterns in historical network data (sometimes called a “training set”) and the implications of that network data on performance or reliability, and then create an program that finds statistically similar patterns in future data. In general, such a program can classify previously unseen data (where the outputs are discrete) or compute a continuous function. In general, we want to use this technology to remediate existing faults or predict future events.
“Big” and “real-time,” when added to analytics, just change the amount of compute cycles needed to get to that output. When I need to analyze more data, I need more compute. When I need to analyze data faster, I need more compute.
Machine Learning, on the other hand, represents a fundamental new model – rather than creating a program to produce an output based on the data, in the case of supervised learning, we use the output to produce a program based on the data. Unsupervised learning, on the other hand seeks to learn progressively more abstract representations of the important features of the data set. This kind of “representation learning” is a subtle, but significant difference: machine learning can be used to find patterns in our data, even (or especially) when we don’t know such patterns exist. Machine learning is rapidly becoming a tool to solve complex pattern recognition problems in networking, Internet of Things, spoken words and imaging.
Come join my session at Mobile World Congress where I will explore the topics of mobile data analytics, discussing use cases where today’s technology is valuable for driving new business and operational models, and a shallow dive into how machine learning will transform the mobile networking industry.