Machine Learning in Data Centers - Not a walk in the park

By CIOReview | Friday, June 1, 2018
116
176
30

Artificial Intelligence merged with machine learning revolutionized the cyber world with many promising abilities and increased efficiency in the data centers. The digital world is predicted to increase the number of machine learning pilots and double the implementation every two years. The data centers can be fully automated, thanks to machine learning algorithms which can run against log data generated by servers, firewalls and routers – any data that has the ability to compromise the data center.

As beneficial as it seems, everything comes at a price. With the constant growth of big data, integrating machine learning into a data center design comes with its fair share of problems. To analyze any corporate data, all the operational logs have to be stored and managed. Moreover, the corporate compliances mandate 3 years of data storage, and with the huge amounts of operations data generated, the log data that provides input to the machine learning systems dwarfs the application data. The interpretation of machine learning algorithms poses a challenge in determining the effectiveness of the algorithm before implementation.

For machine learning to successfully integrate with a data design, algorithms need training data to operate – lots of training data. Apart from data requirements, machine learning finds it hard to cope with non-differentiable discontinuous loss functions which hinder sparse representation detection in an operating data center. Machine learning algorithms are not always guaranteed to function as expected in every case imaginable; failure is part of the package. Hence, the problem at hand must first be thoroughly understood and analyzed in order to apply the correct machine learning algorithm to the data design.