Industrial IoT Edge Architecture for Machine and Deep Learning
events are going to be routine without any anomaly .
Figure 2 shows cost of compute vs data transfer cost benefit of edge-based ML / DL applications .
Learning consists of the following types :
1 . Unsupervised where we do not require training data and assume that normal instances are far more frequent than anomalies .
Figure 9 : Cost of compute vs . data transfer cost benefit of edge-based ML / DL applications
In the following sections , we shall describe the following – a short summary of Deep Learning , edge tier innovations for DL , Platform tier innovations for DL and a use case of sensor based trucking .
WHY MACHINE AND DEEP LEARNING
The science and practice of artificial intelligence is vast and consists of the following types : ( 1 ) Data , ( 2 ) Learning and ( 3 ) Algorithms . Data consists of :
1 . Structured data such as time series data , events and graphs and
2 . Unstructured data such as video , images , speech and text .
2 . Semi-Supervised where training data has labeled instances of only one class . We also assume that normal instances are more frequent than anomaly classes .
3 . Supervised where training data has labeled instances for normal and anomaly classes . We also assume accurate representation of the labels for the classes .
4 . Reinforcement where rewards are provided to guide the learning through policy .
We next consider the types of algorithms . Here we diverge into Machine Learning algorithms which consist of anomaly
IIC Journal of Innovation - 61 -