Summary 

In this chapter, we revisited a previous problem (sketch recognition) but used a different dataset and different approach. Previously, we tackled the problem using CNN, but in this chapter, we identified nuances of how the data was collected, which in turn allowed us to take a different approach using an RNN. As usual, most of the effort was spent in preparing the data for the model. This, in doing so, highlighted some techniques we can use to make our data invariant to scale and translation, as well as the usefulness of reducing details of the inputs (through simplification) to assist our model in more easily finding patterns.

Finally, we highlighted an important aspect of designing interfaces for machine learning systems, that is, adding a layer of transparency and control for the user to help them build a useful mental model of the system and improve the model through explicit user feedback, such as corrections. 

Let's continue our journey into the world of machine learning applications...