Ascribing a timepoint to a particular condition?

[MS Al-Rawi]

Hi 

I've been facing this question at many occasions:

The classifiers tested require training sets. How were these data derived, bearing in mind that it is impossible to ascribe a data point to a particular stimulus given the non-linear, spatially variable nature of the haemodynamic response?

It seems a legitimate question! Does what we get with classifying patterns of cortical activity fulfills an answer to this question? Or there is a more logical answer? 

Any help would be appreciated,

Regards,

- Rawi

[J.A. Etzel]

Given our current knowledge of the BOLD, fMRI resolution, etc, the best
way to handle this seems to be to ensure that the uncertainty will only
make classification accuracy/significance WORSE, not better.

For example, the timing of events in many designs does not guarantee
that the BOLD signal from the prior event will be fully absent from the
following event. But, if the order of events was properly randomized,
and if the partitioning was carefully chosen, you can sometimes make a
convincing argument that any residual signal would only hurt
classification performance, not create a spurious result.

The timing and randomization of events, means of temporal compression,
partitioning scheme, etc. should be described, along with why it guards
against spurious results. You could also describe any tests you did to
make sure everything was ok (label randomizations, different temporal
compression methods, etc.).

Jo

On 9/12/2011 9:00 AM, MS Al-Rawi wrote:
> Hi
>
> I've been facing this question at many occasions:
>
> The classifiers tested require training sets. How were these data
> derived, bearing in mind that it is impossible to ascribe a data point
> to a particular stimulus given the non-linear, spatially variable nature
> of the haemodynamic response?
>

> It seems a legitimate question!Does what we get with classifying

> patterns of cortical activity fulfills an answer to this question? Or
> there is a more logical answer?
>
> Any help would be appreciated,
>
> Regards,
> - Rawi
>

> --
> You received this message because you are subscribed to the Google
> Groups "Princeton MVPA Toolbox for Matlab" group.
> To post to this group, send email to mvpa-t...@googlegroups.com.
> To unsubscribe from this group, send email to
> mvpa-toolbox...@googlegroups.com.
> For more options, visit this group at
> http://groups.google.com/group/mvpa-toolbox?hl=en.