he did a "few" things...
(fame is something different)

Thanks! But I was looking for something for PhD students applying for postdocs. Writing your fist research statement can be quite confusing.

how it feels to do physical activity after the age of 35

N >=2 (max 3) is too complex for the human brain :-)

For the topological/probability work one probably needs PhD level understanding of differential geometry and a lot of probability theory. Considered a basic introduction...
"An Elementary Introduction to Information Geometry"
by someone else; Amari doesn't write basic...
www.mdpi.com/1099-4300/22...

I read this in graduate school when I was sharper and probably got < 20%.
ieeexplore.ieee.org/document/58324

Amari's work is a little like Grossberg's, there's such an immense body and it evolved a lot but let me see if I can come up with something.

Read Perich! (lots of others too)
www.nature.com/articles/s41...
Or his piece in The Transmitter.

oh, I need to learn Poisson Latent Neural Differential Equations (PLNDE) and modify it for fMRI which is what I use mostly (sorry).

Neuroscience got too enamored with *very low dimensions". Let's learn to explored structured higher dimensional spaces... (Amari to the rescue?)
Let's use very low dimensional figures as illustrations of some of the properties that are present in the high-dimensional space, not as the result itself.

The paper does refer briefly to avoiding "homuncular solutions" to brain processing and proposing instead strategies to think in terms of distributed processes.

That would be wonderful!

I sent to the cmu email I found online, hope that's correct.

But at least you have a way that you don't have to discard people that have few instances of a given trial type (e.g., error trials, super slow trials, etc). But this person is not really contributing much at all to the estimate of this trial type.

... the estimates from other levels depending on how the model performs "pooling". So the estimate based on 2-3 trials basically is the mean estimate across participants. While some people think this is great, I don't think there's magic with data. You need lots of data to estimate things properly.

My take is that it doesn't change things much at all with respect to the number needed for standard analysis. The main issue is that in multilevel Bayesian models, depending on how/what you model, you can estimate even with 2-3 trials per condition. But what that really means is that you "inherit"

Thanks, this is super important!