Specific neural network architectures are designed for specific problems. It doesn't mean that one architecture is better than another one—it just means it is better at a specific task.
In this section, we'll be looking at our specific problem, generating music, and see why RNNs are well suited for the task. We'll be building our knowledge of neural network architectures for music throughout this book, by introducing specific concepts in each chapter.
For music generation, we are looking at two specific problems that RNNs solve—operating on sequences in terms of input and output and keeping an internal state of past events. Let's have a look at those properties.
Musical score prediction is analogous to generating music. By predicting the next notes from an input sequence, you can iteratively generate...
