- Need to find a better model than Resnet18
- Keep trying different models and parameters to maximize 6-region classification accuracy -45%, 100epochs, around 60% should be good (classifying 6 regions)
https://www.brainmeeting2021.com/en/event-submissions
- Add figures
- Update it based on progress
- Methods section
- Read articles found to get a better idea on how to write intro
- Run captum analysis on 45% - can't run one analysis but rest are fine
- Implement neuron attribution from CAPTUM -have to build the pipeline, no tutorial with a smooth transition
- https://pytorch-lightning.readthedocs.io/en/latest/extensions/metrics.html#classification-metrics
- https://forums.fast.ai/t/determining-when-you-are-overfitting-underfitting-or-just-right/7732/2
- https://github.com/PyTorchLightning/pytorch-lightning/blob/master/pl_examples/basic_examples/conv_sequential_example.py
- https://github.com/pytorchbearer/torchbearer
- Work on writing a program to do descriptive statistics (e.g. range, mean, stdev) that can describe potential differences between the different classifications of images (visp, visal, visam, visrl, vispm, visl).
- Work on another shuffling method of data --bootstrapping in which we can also get the mean and stdev of its results once we have 10 results from the 10 different trials that were ran.