a way to assess noisiness in training data #7
Description
Some assumptions that we share are:
- for similar tasks, regardless of visual stimuli, activity in the brain should fire on roughly the same channels at roughly the same rates,
- @Sentdex is storing consistent thoughts only,
- and that the device is sensitive enough or that these thoughts occur in the same region
Having consistent brain activations is probably hard for any task. Maybe part of the dataset collection process should train the human as well to ensure that consistent data is collected. If something isn't working, think of something specific until it does. Maybe "THINK LEFT" just isn't interesting enough to @Sentdex 's brain, so "IMAGINE EATING A PB+J" might work as a better proxy.
Something like https://github.com/cgnorthcutt/cleanlab/ could help with the user interaction piece.
Dunno. Just thinking out loud here.
-Russell