Make Discussion section more robust.

[praeducer]

• [ ] Add thoughts of how this fills in the gaps of scientific research.
• [ ] Summary of all results. Add more about collaboration and data accessibility.
• [ ] Make a jumping off point for open research. Where we can take the fielding the future.
• [ ] Allows reproducibility and replicability in our field.

[praeducer]

Make sure to cite this paper and address it's concerns:

4.2.4 Alternative approaches

Some research initiatives and databases have attempted to develop an alternative approach to sharing human subject data to better protect subjects' privacy or to respect subjects' autonomy on how their data should be used. For example, instead of sharing raw individual-level data, the Enhancing Neuroimaging Genetics through Meta-analysis (ENIGMA) provides analysis protocols to run at a local site and perform meta-analysis on results (e.g., summary statistics) returned from the local sites (Thompson et al., 2014; Thompson et al., 2020). The Collaborative Informatics and Neuroimaging Suite Toolkit for Anonymous Computation (COINSTAC) is another large-scale research consortium that offers tools for decentralized analysis which allows researchers to conduct both meta- and mega-analysis without pooling the data at one central place (Ming et al., 2017; Plis et al., 2016). Groups of users run common analyses on their local sites using their own data and the results of these analyses are then synchronized to the cloud and undergo aggregate analyses processes using all local data. The federated computing of COINSTAC also provides heightened privacy protection using advanced statistical algorithms, such as differential privacy, and this model may open up a way to gain access to data that researchers are unable to share due to local regulatory restrictions (Ming et al., 2017; White et al., 2020). However, it is important to note that federated computing could make data sharing more burdensome because it would require significant synchronized agreement efforts among all the sites to run the analysis locally or to change the protocol.

https://onlinelibrary.wiley.com/doi/full/10.1002/hbm.25803

[krootesmurdy]

I've made some edits and tagged some still need items. But you can send it off to Vince as is. I'll continue working on Discussions and Results next week.

[praeducer]

Limitation section

• Federated learning itself has some limitations:
• Communication speed bottlenecks.
• in some cases it can be possible to re-create source datasets using only derived results