Reliability
The project is to study relationship between reliability and predictability
Results
- theory
- def of R
- R bounds predictive accuracy for any subsequent task
- Rhat --> E(R)
- simulation
- Rhat --> E(R) in practice for some (eg, gaussian) simulation
- Rhat provides a more useful bound than ICC or I2C2 for a variety of simulated settings
- we can use Rhat to choose the most discriminable parameter (eg, threshold)
- real
- optimal reliablity yields optimal predictive accuracy
- best pipeline = product of marginals
- thresholding vs. binning ranks
- which atlas/resolution
- DTI vs. fMRI
@jovo make each of the above points into a sentence.