I want to examine the neural response to stimulus in behavior session. For the session, I have 50 trials. If I want to detect the neural response trial by trial, for some trials, I have no spikes during baseline, which means I couldn't normalize my data with z-score. Anyone has any suggestions about how to do normalization?
Hi Tianyang,
If my understanding of your problem description is not erroneous, a possible solution consists in -adaptively- normalizing the targets used in learning. Please check for this reference: https://arxiv.org/abs/1602.07714
Hoping it will be helpfull
Hi Tianyang,
Basically, if you have a clean baseline, then you could use it to rescale your data. Thus, you may apply either "Decibel conversion or "Percentage change" for this purpose.
HTH.
Samer
Thanks you, Stephane!
I checked the reference, which seems not fit with my question. They are quite abstract for me.
Dear Tianyang,
My previous post referred to machine learning and artificial neural networks, i.e. wrong understanding of your problem context.
Thanks to Samer Sarsam's answer, I suggest that you follow this here-after reference which might help you to solve your problem in an event-related potential model application context: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3183439/pdf/fpsyg-02-00236.pdf
Regards
Dear Tianyang Ma,
look the link.
http://howto.commetrics.com/methodology/statistics/normalization/
Regards, Shafagat
Hi Tianyang,
Sorry for the delayed response.
Single-trial baselining is a tricky issue, best to avoid it. You could try a linear baseline subtraction for the single-trial analyses, although this would preclude using dB or percent change at the trial-average level.
However, the method I suggested previously can be used if you have continuous data only with clean baseline.
Finally, I highly suggest reading the article recommended by Stéphane Breton.
HTH.
Samer
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence