banded ridge regression

[alireza-kr]

Hello

I am curious to know your thoughts on using banded ridge regression (La Tour et al., 2022) to disentangle correlated response estimates between trials that are close in time.

• La Tour, T. D., Eickenberg, M., Nunez-Elizalde, A. O., & Gallant, J. L. (2022). Feature-space selection with banded ridge regression. NeuroImage, 264, 119728. https://doi.org/10.1016/j.neuroimage.2022.119728

[kendrickkay]

In GLMsingle, there is only one set of predictors that sort of "group" together, namely, the single trial predictors. (The other predictors, nuisance/drift/etc., are fitted using least squares up front.) So, banded ridge regression doesn't seem to apply in any straightforward way... did you have something in mind?

[alireza-kr]

Thank you Kendrick for your answer.

I read in your paper [1] about the instabilities in GLM estimation that can arise from having correlated single-trial predictors. You suggested using ridge regression to address this. However, ridge regression applies the same regularization hyperparameter to all features. What I suggest is using banded ridge regression, which allows for a separate regularization level for each feature.

Prince, J. S., Charest, I., Kurzawski, J. W., Pyles, J. A., Tarr, M. J., & Kay, K. N. (2022). Improving the accuracy of single-trial fMRI response estimates using GLMsingle. eLife, 11. https://doi.org/10.7554/elife.77599

[kendrickkay]

Yes, but are you suggesting a different regularization level for every single trial? That seems a bit crazy

…

On Jun 28, 2024, at 6:45 PM, Alireza Karami ***@***.***> wrote: Thank you Kendrick for your answer. I read in your paper [1] about the instabilities in GLM estimation that can arise from having correlated single-trial predictors. You suggested using ridge regression to address this. However, ridge regression applies the same regularization hyperparameter to all features. What I suggest is using banded ridge regression, which allows for a separate regularization level for each feature. Prince, J. S., Charest, I., Kurzawski, J. W., Pyles, J. A., Tarr, M. J., & Kay, K. N. (2022). Improving the accuracy of single-trial fMRI response estimates using GLMsingle. eLife, 11. https://doi.org/10.7554/elife.77599 — Reply to this email directly, view it on GitHub <#139 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAU2DEQIB5RGOFX3FYXPR4LZJXYQ3AVCNFSM6AAAAABJ2WEDROVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TSMBZGUYTC>. You are receiving this because you commented.