Commit 00d906fc authored by Ying-Qiu Zheng's avatar Ying-Qiu Zheng
Browse files

update

parent 5ba3eeb3
### Weighted average coefficients of all subjects (but the new subject) to do the prediction
![](figs/all_vs_best100_reg.png)
## Summary
### 1) What is the set of bases that can better represent task-activation in individuals? - how to find a better set of bases? - more advanced methods may come in...(factorise resting data into modes..)
* for now we focus on the comparison of simple methods, i.e., ICA, PCA, Laplacian eigenmaps (with number of components predetermined)
......@@ -21,7 +18,12 @@
* so it is probably biased by group level effects...
##### using residual bases to make prediction also improves prediction accuracy
![](../2020MAR24/figs/model_accuracy.png)
* How to further improve the performance?
* How to further improve the accuracy? - improve bases and coefficients used in the prediction
* Use linear combination of other subjects' coefficients (e.g. weighted average) to do the prediction - seems improvement is very trivial
![](figs/all_vs_best100_reg.png)
* Or learn a new transformation (under some contraints) to apply to the coefficients or bases
* learn the relatedness of different tasks so that shared information could be used (coefficients are low-rank?)
* Apply a prior to the coefficients (or the bases)... and incorporate a Bayesian approach
### 3) Amplitude prediction
![](../2020MAR24/figs/amplitude_prediction_hcp.png)
......
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment