In the preceding seven chapters we have discussed generalized forms of structure in neural network representations, built models to respect this structure, and studied the ensuing relationships between such models and more canonical forms of structured representations in machine learning, such as analytic equivariance. We have further leveraged these models to test hypotheses from theoretical and computational neuroscience, providing empirical support for some while simultaneously introducing novel methods to the machine learning community in the contexts of both structured representation learning and structure discovery. In conclusion, we return to our original motivations outlined at the start of this book, and see how far we have come towards realizing them, while simultaneously looking at the future directions illuminated by this book.