Using this dataset:
As series of experiments to see what model, loss, optimalization combo will work best. Mostly for exercise and personal knowledge rather than something strictly scientific, especially since the dataset is too small for my liking! :)
Seeing as the dataset is not as big as it should be for some "serious" application, the project once finished will be moved on to be used on a different, bigger classification set.
To Do:
⬜ Purely implementing
:white_check_mark: "Basic" CNN
:white_check_mark: "Basic" CNN + LeakyReLU
:white_large_square: Config file and arg pass for easy pick of parameters, instead of hardcoded
:white_large_square: Custom range of Learning rate
:white_large_square: Loss functions:
✅ CrossEntropyLoss
⬜ BCE
⬜ Focal Loss (also an idea: test with a dataset with a class that has more advantage)
⬜ Mutual Information?
⬜ Expand this list, ha!
⬜ Research
:white_large_square: Browse through publicly available papers on new methods, such by M. Ravanelli.
⬜ Tests
⬜ Additionally, add results of each model-optim-loss combo here :)
Other project ideas, before I forget 💀 :
Wasserstein Loss & GAN