Regarding conflicts of interest, confidentiality, anonymity, ethical guidelines, commitment, respect and scheduling please refer to the guidelines provided by the MIR community on the dedicated conference websites:
For technical guidelines on deep learning and music you can use this humble following advices. Check for completeness of details about:
-
Music aspects
- datasets used, please refer to datasets.md
- data augmentation (Pitch shift, Time-stretch, Mixing, Circular shift, Noise addition, Filter, Dropout, ...)
- input type (Raw signal, Time-frequency representation, ...)
- number of dimension used as input (1D, 2D, ...)
-
Deep learning aspects:
- architectures, please refer to architectures.md
- learning rate (Fixed or changing and range)
- framework, please refer to frameworks.md
- reproducibility, please refer to reproducibility.md
- activation function (ReLU, Leaky ReLU, Sigmoid, Softmax, ...)
- number of epochs
- batch size (the bigger the better but generally between 16 and 150)
- loss function (RMSE, Cross-entropy, ...)
- number of layers
- dropout ratio
- cpu or gpu usage and description
- computation time (Global or per epoch)
- optimizer (Adam, SGD, ...)
-
General aspects:
- source code provided
- description of the task similar to existing ones
- citing relevant literature from dl4m.bib