PhD Doctor of Philosophy candidate
Robot Ed Sheeran: Making an AI Composer
Current state-of-the-art ML for music (see: OpenAI, Google Magenta) is bad at learning long-form structure. I believe this is because their approach of swallowing and vomiting MIDI files does not reflect the vast amount of prior information we have as humans: we have ears that pick up on the subtle ways that pitches interact, and we're hard-wired to understand linguistic structures. Let's build a better AI with these inductive biases and generate some funk!