Largest artificial neural network?

I have another question: who has most advanced AI to this day (sentient and not) and the second question is: does quantum computing equals to rapid AI developement?

Can you write something more about ANNs being just a geek meme? Is it just a failed experiment, are they just not very efficient and what solutions are better than ANNs? Does simulating human brain has even a sense to achieve revolution (sorry, I don't know how to say that in other words) in the AI sector.

>ANNs is a fucking meme, brought to life just to produce more papers
I don't think you see the potential. If the computing pwoer were large enough we could surpass human inteligence quickly. In a lot of ways we already did.

Imagine the entire human brain working on a single mathematical problem, not being distracted by emotions, or videos of cats or porn.

Anyone wanna answer this? I've wondered this myself. Hopefully quantum computer will become optimized enough to where to we start working on this and neural networks.

Neural networks completely blow everything else out of the water in computer vision and speech recognition. Sure dumb undergrads make them do memey stuff for upboats but that doesn't take away from the fact that they're really good at stuff we previously didn't have good solutions for.

DARPA

How many tendrils do neurons have?

Why is this so complicated? Is it because neurons are all synchronized?

Each cortical neuron can make up to around 1,000 connections.

>synchronized?
Huh? I'm not sure what you mean, but I'm pretty sure the answer is no.

Google up how neurons work.

1000? That's a lot of processing

>synchronized
wasn't it recently found that dendrites also process information, instead of just acting as buses?

nature.com/nature/journal/v503/n7474/full/nature12600.html

or is this bunk?

>A novel hardware architecture will be needed.

What is turing complete? just simulate the "hardware" of the neurons virtually.

>simulate the hardware to solve latency problems
are you literally retarded?

Just add more real life hardware to solve the latency problems. The point is you don't need special or novel newley developed real life hardware to simulate a mind.

If you want to make it as small as the human brain you might have to make special real life hardware.

I don't think simulating interactions between nerve cells is viable with anything that's centralized, especially X86, ARM, or even RISC architectures. GPU-like cluster computing is better, but still not there.

All you need is a massive network of small, simple, more or less fixed function chips connected together through high bandwidth buses (as it makes sense). These units can perform different macro tasks, it really depends how you want to lay it out.

Ideally you have truly novel hardware. Something using memristors, so you control state by the very nature of the machine's operation.

I was thinking you would just have a neural network on normal hardware, honestly I don't see the problem with it. Make it so any number of connections can be made throughout the "hierarchy" as in row three can connect to row one not just row two and so on. I can't get into detail but I don't see why you can't simulate the physical architecture of biological neurons virtually.

It's due to the way modern hardware and memory works. Just get into the basics of branch prediction, prefetching, and inevitability of cache misses, begins to show why something like this will be far too slow on traditional, general computing hardware.

It needs an architecture made for it.

> How big is the biggest neural network that wed have ever built

About the size of OP's mom

>how does it compare to the human brain?

Pretty small, considering the size of OP's mother's brain.

>Also, how far can we go in terms of building a large neural network?
The only real limit is currently cost, since adding processors and hooking them up gets expensive once you get into the supercomputer range.

Well I don't know enough to suggest or contribute ideas to the design of this new architecture.

It needs to solve what problems exactly? cache misses, branch prediction and prefetching issues?

That sounds like a job for the FPGA.

>memeristors
kek

R=I/(420-M)

Where I is the current and M is how dank the meme is

More or less, nerves aren't so much about computation, they're about state. The mind's overall activity is about the state of the whole, relative to what the parts are communicating, and to where, in accordance with their state. What defines their state? Their inputs and properties. Take a small part of the visual processing regions of a rat's brain. If you watch a given cluster of cells, their axons, and their dendrites, you can see how they turn on and off inputs, and thus alter their overall state when shown a rotating gradient, something with hard contrast, whatever. Further, look at how pyramidal cells encode their outputs based on their inputs when eg complex object recognition.

The hardware needs to do very simple tasks that change state, or operate relative to state, repeatedly. And these clusters need to communicate with each other very efficiently. They also need to be capable of simulating complex states, like bonding affinity of a given compound to a subunit of some receptor, neurotransmitter metabolism and reuptake, etc. If you chose to go that route.

This is why centralized computing doesn't work, even in you run through memory sequentially by design (eliminating cache misses), it's just too slow. You're designing to work around hardware limitations instead of just running it on something more suitable.

>Take a small part of the visual processing regions of a rat's brain.
> They also need to be capable of simulating complex states, like bonding affinity of a given compound to a subunit of some receptor, neurotransmitter metabolism and reuptake, etc. If you chose to go that route.
I think there's some severe misunderstanding going on in this thread. I'm pretty sure that when people are using ANNs for things like computer vision, as opposed to studying braings, they aren't simulating actual biological neurons.

I'm not entirely sure how did I manage to accidentally type 'g' between 'n' and 's'.

I'm talking about simulating actual nerves, but optimizing such that you don't have to go to the point of simulating ion channels on a particle level.

Even in the case of another neural net approaches, novel hardware is probably the best approach. There are companies out there building large clusters made of units with some number of cheaper ARM processors.

I'm making some pretty terrible grammatical errors myself. My neck snapped around earlier and I'm feeling that awful mix of anxiety and pre-migrainous sensations. Numb hands, reduced awareness, etc. Not sure what causes it, mechanically, as it's fully reversible. Could even be entirely psychogenic.

Kind of want to kill myself tbqh senpai :^).

Never mind, I figured it out. Realized I hadn't drank anything in ~36+ hours, and had just eaten some chocolate and turkey (which is quite salty). It dilated blood vessels and acted as a diuretic, further disrupting electrolyte and water balance. Beginning to feel better.

Off topic, but whatever. I don't give a fuck and never will.

Nope. It's definitely something migraine-esque.

I recently had a long period of sleep deprivation. Woke up feeling fine enough. Have neck problems, moved my head wrong relative to the position of my shoulders, heard and felt a large snapping sound. Progressively felt shittier and shittier. Also have jaw problems.
-Can't keep anything in memory very long / well.
-Poor awareness of context, beyond the awareness that I'm poorly aware.
-Forearms and hands feels numb
-Legs feel strange and as though I'm floating
-Can't see very well. All parts of the visual fields are present, no tunnel vision or some such. Everything has an appearance of extreme contrast, and colors are vivid. I feel as though I can see, but do not understand much at one time.

I don't think it's a compressed or kinked blood vessel. I don't have any idea what would cause this or if it even relates with the neck snapping (but has correlated in the past). I already posted before, so I figured I'd ask, does this sound like anything to anyone?

10 neurons? You mean 1 layer of 10? Thats not true, just 10 can be simulated even in my computer

Why they need 150 layers? For every problem I tried more than 3 layers is never more usefull than 3, what kind of problem requires 150?

Right now quantum computing is useless, there are only 3 problems that can be solved by quantum computing, you can't make every program in a quantum computer as people use to think

Read subsequent posts.

slapping more trasistors doesn't automatically make CPU faster

yeah with 3d transistors duh

Yeah that's been known for a while now, it's a shit show.

Literally every month a group will publish a paper about how they discovered a new way that neurons encode information or demonstrate a new kind of plasticity

schizophrenia

unless you're the same idiot that smashed his head and didn't go to the doctor, in which case you're still an idiot and should still go to see a doctor, like perhaps 10 people told you last time.

Oh no :(

I'm not.

Thank goodness he was unbearable.