ABSTRACT
An artificial neural network is one of the greatest findings in the computing world. Its applicability is found across the real world applications such as text processing, vision, medicine, and forecasting. A network is essentially computational machine which learns from its experience and tries to mimic the behavior of biological brain. The learning of a network happens through the feedback. Since cloud workloads are very dynamic and their pattern gets changed frequently, a typical neural network may act unreliably in a long run. Therefore, a frequent training is required for a reasonable performance over a long period. In this chapter, we will discuss an alternative learning mechanism which makes a network robust. In this approach, a network keeps receiving continuous feedback on its performance which is utilized to improve the next forecast. Since the network is able to capture the dynamic patterns in the workloads through its continuous feedback, the learning scheme is referred to as self directed learning. Furthermore, an improved blackhole learning algorithm is also discussed which divides the population into sub-populations and local information is incorporated in the position update procedure for a more diverse population. The performance of a non directed learning and self directed learning based predictive models is studied and analyzed critically through a series of experiments.
