Brain Mapping

https://www.technologyreview.com/s/609070/inside-the-moonshot-effort-to-finally-figure-out-the-brain/
Here's a report on a study to map a 1 mm cube of a rat brain, with an estimated 100K neurons and 1B synapses in it, in so much detail that the scientists can see the organelles inside the individual cells. As the article explains, it's a painstakingly slow process right now, but the researchers are developing tools that they expect (just as with the Human Genome Project) will greatly speed up the process over time. 1 mm now, maybe 2 or 4 next time...

One point from the article is that the study isn't just about doing a detailed map of this meatcube. It's about figuring out how the heck learning works, too. How is it, the authors ask, that a manmade neural network needs a ton of training material to recognize a dog but a human kid can figure it out from one example? (I think a partial answer is that the human is seeing many individual animation frames, but that's not all.) One thing not mentioned here but that I'm curious about is the notion of... I think it's called "rate encoding". That is, we've kind of assumed that rapid neuron firing = more activity, but what if the exact timing of the pulses matters so that 1101 is different from 1011?

---

One other thing: a plug for the new audio edition of the first "Thousand Tales" volume, and a note that one of the other stand-alone volumes, "Extra Lives", is now permanently free on Amazon and Smashwords!

Comments

[Mix-up]

glade to see you again and it was an interesting article, I hope yo see more of you again in the future soon.

[mastermenthe]

yay

[book_burner]

Always great to see you on the site!

One point from the article is that the study isn't just about doing a detailed map of this meatcube. It's about figuring out how the heck learning works, too. How is it, the authors ask, that a manmade neural network needs a ton of training material to recognize a dog but a human kid can figure it out from one example? (I think a partial answer is that the human is seeing many individual animation frames, but that's not all.) One thing not mentioned here but that I'm curious about is the notion of... I think it's called "rate encoding". That is, we've kind of assumed that rapid neuron firing = more activity, but what if the exact timing of the pulses matters so that 1101 is different from 1011?

Look up "predictive coding" and the "Blessing of Abstraction".

[KrisSnow]

4735221
Thank you!
As for the article, the "predictive coding" looks interesting. Particularly this bit:
the expected precision of visual prediction errors likely varies between dawn and dusk, such that greater conditional confidence is assigned to errors in broad daylight than errors in prediction at nightfall...
As though humans were instinctively predisposed to see night as a time of uncertainty and mystery for a basic neural reason, even beyond the reason that it's when some predators come out.

[KrisSnow]

4734181
Thanks! That cheers me up a bit. I don't really know what to post, though! Feels like I've kinda exhausted my FiO ideas.

[book_burner]

4737300
Not so much instinctively predisposed as, our eyes just work better during the day. Owls might be more predisposed to think of nighttime as a good, clear, understandable time, with daytime being this horrifying noisy glare.