From 2010, the ML technique DL became an important new method for implementing neural networks (NNs). Thanks to the application of a hierarchical distributed representation across multiple layers, NNs have been restored to their former glory.

In the I-IoT, the most common NN is the recurrent neural network (RNN) or the long short-term memory (LSTM) network. LSTMs are normally used to predict time-series data. Recently, however, we have started to use Convolutional Neural Networks (CNNs) in the IoT sector as well.

Several companies, including Google and NVIDIA, have been developing new hardware processors to support this technology. A few vendors have also applied GPUs to NNs, thanks to the ability of GPUs to work with connected graphs. Google has also developed a new ASIC processor called a tensor processing unit (TPU) and the TensorFlow library to work with GPUs or TPUs either on-premise or in the cloud. Later, we will look at an example of using RNNs with TensorFlow.