Maqenta
Maqenta copied to clipboard
Generating music using quantum machine learning models. (QuGAN and QLSTM)
Maqenta
Magenta, but in Quantum.
Roadmap
-
Learn about GANs
- [x] Watch Ian Goodfellow: Generative Adversarial Networks.
- [x] Watch Seth Lloyd: Quantum Generative Adversarial Networks.
- [x] Go through PennyLane's Tutorial of QuGANs.
- [x] Chris Olah - Understanding LSTM Networks
- [ ] Go through GANSynth tutorial on Magenta.
- [ ] Port GANSynth to Quantum.
-
Learn about RNNs & LSTMs
- [x] Implement Quantum Long Short-Term memory in PennyLane.
- [x] Implement a music generation algorithm using the above.
- [ ] Go through RNN music generation tutorial on Magenta.
- [ ] Port the above to Quantum.
-
Learn more about other QML Methods that could be possible for generating music.
-
Learn Music Theory to be able to find even more ideas for generating Quantum Music.
-
Read these papers
-
Future Development.
- [ ] Develop the code for EQGAN.
- [ ] Change the measurements from global to local in the QuGAN module, because it gets trapped inside a barren plateu very quickly.
- [ ] The Issue of monotony (same offset between notes).
- [ ] Using a tunable vector
w
to be inner producted with the output of the functions, basically adding a classical layer that needs to be optimized. - [ ] Cross-check to see if there's any difference in tuning a matrix to generate multiple next notes or just a vector to finding one next note.