/sci/ - Science & Math

>>9306572
Wat?

There is no NN I've heard of that reached even a percent of the size of the brain (in number of neurons). The human brain has about 90 billion or so, what do NN's for scientific / industrial applications use .. dozens at best? Do you even know what you're talking about?

>>9306572
>You'd think it'd be pretty important thing to simulate
Nope.
>Purpose: Long-term neuropsychological outcome was studied in 71 patients who underwent hemispherectomy for severe and intractable seizures at The Johns Hopkins Hospital between 1968 and 1997 and who agreed to participate.
>Methods: Patients and caretakers were interviewed, and patients were administered standard measures of intelligence, receptive and expressive language, visual-motor skills, adaptive/developmental functioning, and behavior.
>Cognitive measures typically changed little between surgery and follow-up
http://onlinelibrary.wiley.com/doi/10.1111/j.0013-9580.2004.15303.x/abstract;jsessionid=D4AC787F0C3FD4F5C4D3E01457E79D71.f03t01
>most, if not all, animals have two brain hemispheres
You have two kidneys too, that doesn't mean there's any sort of additional functionality provided by an interaction between those two kidneys that wouldn't already be covered by modeling the functionality of one kidney.

>>9306581
What are you trying to make a point about here? OP's wrong, but what he's wrong about has nothing to do with whatever inane issue you think you're bringing up.

>>9306593
Neh, I think I have a point. The human brain's hemispheres are obviously something inherent to brains or constructs of similar complexity. If you're sticking together 10 neurons for image classification, what would you gain from separating them into hemispheres?

>>9306602
parallel processing

>>9306602
I just proved you didn't have a point though. You can remove one hemisphere from someone's brain and the evidence we have suggests there isn't any significant decrease in cognitive power following that hemisphere excision.

>>9306585
>Cognitive measures typically changed little between surgery and follow-up

>You can remove one hemisphere from someone's brain and the evidence we have suggests there isn't any significant decrease in cognitive power

That paper does not prove that at all. The seizures render those areas of the brain that are removed as disfunctional and from it as your paper shows, the mean IQ of these groups are extremely low. It shows removing the damaged part of the brain does not negatively effect them but the damaged area did significantly decrease their cognitive abilities. As such, if you removed a healthy child's part of the brain you could still expect a decrease in cognitive abilities like these children have. It of course depends on to what degree and how old the child is but no such experiment has been done to shows there would be no decrease in ability.

Long story short, removing a damaged part of the brain and removing a health part are two very different things.

I honestly agree with your overall premise that mimicking this system in neural networks would be unnecessary for most systems but the premise you can live just as well with one hemisphere is silly. There is a limit to neural plasticity even in children.

>>9306572
that's not true, GANs have 2 networks, capsNET kind of have 2 although not at same time, synthetic gradients technically have +1 for each layers to predict the gradient. Also many large scale projects like Siri, alphago etc use multiple neural network each with it's won task.

They don't behave like hemispheres per say but we do have multiple-network systems, it's just that nobody has any use-case for something like what the left/right hemispheres do yet.

>>9306581
thousands