Python Deep Learning - Second Edition

DL in the Cloud

In this chapter, we are discussing a serious topic, AVs and how to apply DL techniques in them. Let's see how to approach this task in practice. First, let's observe that in deep networks (as with most ML algorithms), we have two phases—training and inference. In most production environments, the network is trained once, and then used only in inference mode to solve tasks. If we obtain additional training data during the course of events, we can eventually train the network again (for example, using transfer learning). Then, we can embed the new model in the production environment until we need to retrain it again and so on. The alternative to this is incremental learning, having the model (network) constantly learn from new ...