🗑️ 🧵 Neuralink Compression Challenge
Anonymous at Sat, 25 May 2024 12:45:33 UTC No. 16192528
You should be able to solve this.
Anonymous at Sat, 25 May 2024 12:52:59 UTC No. 16192538
>>16192528
Why is the out bandwidth so low?
It's to a phone in close proximity should be able to get tens or hundred of mbps
Anonymous at Sat, 25 May 2024 13:21:37 UTC No. 16192571
>>16192538
my first guess would be that having it too high might interfere with brain functioning
Anonymous at Sat, 25 May 2024 13:38:26 UTC No. 16192585
>>16192528
Have they tried zstd man
Anonymous at Sat, 25 May 2024 14:20:54 UTC No. 16192608
>>16192528
Why should this even be possible? If you are using a real time data feed then the encoding chunks really have no way of being compressed. Consider the 1 ms timing. With 200 Mbps you have 200 Kb of data to encode, but there is 1024 data sources. In this chunk, each electrode will have ~200 bits of data that has to be narrowed down to 1 bit.
This is stupid.
What they want is someone to develop a magic electrode model of neural activity.
Anonymous at Sat, 25 May 2024 14:30:17 UTC No. 16192617
>>16192608
>If you are using a real time data feed then the encoding chunks really have no way of being compressed
>bog standard zip achieves 2.2 ratio
just kys already. obviously you can do somewhat better than that, they wanna find out how much.
Anonymous at Sat, 25 May 2024 14:44:58 UTC No. 16192636
homomorphic compression, but the issue will be determination of context, so you'd need to fine a general solution to thoughts themselves?
Anonymous at Sat, 25 May 2024 14:49:29 UTC No. 16192641
>>16192528
idk......have they tried multiplexing yet?
Anonymous at Sat, 25 May 2024 14:53:47 UTC No. 16192646
>>16192608
Antisemite destroys the OP once again
Anonymous at Sat, 25 May 2024 15:07:43 UTC No. 16192666
>>16192528
yeh i haven't looked at the standard yet, but 1024 pins outputting serial data (.wav) should be easy to multiplex into a 10-bit parallel bus. That's 100x compression. Add your .zip or whatever and there's your 200x compression rate
Anonymous at Sat, 25 May 2024 15:37:35 UTC No. 16192686
>>16192666
Could you extrapolate on this?
Anonymous at Sat, 25 May 2024 15:39:23 UTC No. 16192688
>>16192666
The channels are all firing at the same time retard
Anonymous at Sat, 25 May 2024 16:06:55 UTC No. 16192729
>>16192686
other guy has a point, but if you can somehow buffer the 1024 inputs then stagger them, then you'd still be able to transmit the data through 10 channels at 100*200kHz = 20MHz i think
.....but then to reach 200x you'd need 10 parallel .zip encoders lmao
Maybe instead of 1024 electrodes, multiplex together 2048 into 11 bit streams (200x), skip the .zip, and transmit 11 channels at 40MHz thru RF? FCC's gonna love that lel
Yeah i don't think this is the solution. But potentially, a multiplexed encoding could be easier to compress. These probes are all in the same brain region, so their impulses should have high redundancy in magnitude (i.e. binary encoding), right?
Anonymous at Sat, 25 May 2024 16:10:59 UTC No. 16192736
>>16192538
>neuralink electrodes in the brain -> neuralink device on the brain
~200Mbps of data via wires
>neuralink device -> smartphone (or computer)
~1Mbps low energy bluetooth bandwidth for longevity of battery / faster computes
Its not using wifi which uses few watts of power. This 1Mbps is <10mW
Anonymous at Sat, 25 May 2024 16:14:18 UTC No. 16192740
>>16192736
>low energy bluetooth
>wifi which uses few watts of power
I see. Didn't know wifi used that much power over short range.
Anonymous at Sat, 25 May 2024 16:23:31 UTC No. 16192753
>>16192736
Also they likely want to implement some hardware solution on the asic neuralink chip itself for real time compression and not be limited by computes delays/etc. Pure hardware level compression is what they're probably looking to solve for real time processing of signals
Anonymous at Sat, 25 May 2024 16:59:22 UTC No. 16192795
my original idea was to sort the electrodes by signal strength and send the order along with a lazily fitted curve (likely a sigmoid). then you would only need to send small correction terms to make it lossless as well as updates to the order and curve parameters (and electrodes located at the top or bottom of the sigmoid wouldnt even need to be moved around that much).
only problem is the electrodes are 10 bit, so 200x compression would require correction terms of 10/200 = 0.05 bits. sending 20 frames at once (which at 20kHz with 1ms response is the upper limit) would require 20*10/200 = 1 bit correction terms, and you somehow have to reconstruct 20 electrode signals from that. also, the data that the challenge provides seems to be individual electrodes.
Anonymous at Sat, 25 May 2024 17:07:34 UTC No. 16192809
>>16192795
>sort the data and adjust so it fits into a sigma wave and then send that, what incredible compression
Anonymous at Sat, 25 May 2024 17:12:12 UTC No. 16192816
>>16192809
you wouldnt resend the order every time, retard. the signal of individual electrodes is unlikely to drastically change in 50 us, or even 1ms.
Anonymous at Sat, 25 May 2024 17:20:33 UTC No. 16192826
Slightly off-topic, but why wireless?
Wouldn't it be better to sew (or adhere) thin wires through the skin? I'm picturing a zig-zag from scalp to hip. That way, you're not fighting through the impedence of air, or the possibility of malicious actors, and have the option to inductively charge the implant through your hip.
I'm reading that there's some work on silk-based subdermal fibres that could be promising
Anonymous at Sat, 25 May 2024 17:32:24 UTC No. 16192837
>>16192608
There's probably domain specific tricks to achieve impossible compression ratios.
https://lichess.org/@/lichess/blog/
Anonymous at Sat, 25 May 2024 17:54:16 UTC No. 16192864
>>16192826
Anything that's not fully inert produces an immune reaction from the body proportional to its size.
The entire implant is encased in glass.
It can inductively charge through a pillow which is enough, wireless is the right choice.
Anonymous at Sat, 25 May 2024 17:59:30 UTC No. 16192872
>>16192864
https://newatlas.com/implantable-el
>Silicon has not conclusively been proven to be biocompatible, but all studies so far have shown it to be safe.
>The devices also contain gold and titanium, which are required for the electrical connections. Because they are biocompatible but not biodegradable [...]
Anonymous at Sat, 25 May 2024 18:22:06 UTC No. 16192897
What's the prize? Do they want talented people writing free code for billionaires? Kek I can imagine Elon saying, "Thanks for your contribution, you've helped us improve humanity!"
Anonymous at Sat, 25 May 2024 18:24:11 UTC No. 16192898
>>16192897
Is money ur only incentive in life u fag
I hate ppl like u
Anonymous at Sat, 25 May 2024 18:46:32 UTC No. 16192931
>>16192528
What's the current hutter prize compression ratio?
That is lossless too and people have had some success
Anonymous at Sat, 25 May 2024 18:49:04 UTC No. 16192934
>>16192931
>it's 8.88x
yeah 200x is entirely delusional. surely lossless is the wrong approach
Anonymous at Sat, 25 May 2024 18:55:55 UTC No. 16192943
>>16192934
Also wikipedia text is way more predictable than this.
All the files have tons of noise in them.
How the fuck are you going to losslessly compress noise 200x?
Whoever made that requirement needs to be fired
Anonymous at Sun, 26 May 2024 00:28:07 UTC No. 16193376
>>16192943
elon wrote it
Anonymous at Sun, 26 May 2024 00:57:18 UTC No. 16193400
>>16192528
How is this related to neuralink? This is just a compression algorithm improvement which is essentially pure math.
Who wrote this shit?
Anonymous at Sun, 26 May 2024 01:15:34 UTC No. 16193414
>>16193400
AI Wrote it. Notice that it is a prompt to humans that it awaits an answer on.
Anonymous at Sun, 26 May 2024 01:49:12 UTC No. 16193446
>>16192528
I'm doubtful the 200x compression ratio is even achievable, let alone the other constraints. Try Huffman encoding first, and if you can't achieve significantly better than 200x compression then anything else is a waste of time.
If I were smart enough to do this I'd file a patent on the algorithm and tell Elon Musk that he can have it on the condition he either sucks my dick and licks my balls, or buys it from me for a million dollars plus royalties in perpetuity.
What I wouldn't do is post my code on a Mongolian grasshopper forum.
Anonymous at Sun, 26 May 2024 02:06:19 UTC No. 16193461
>>16192898
simp
Anonymous at Sun, 26 May 2024 03:27:24 UTC No. 16193532
Why 20kHz in the first place? Don't the fastest neurons only fire ~200 times a second?
Looking at the first file, the signals being measured (I see two) only seems to be about 70 Hz. Start by decimating the output until the monkey can't play pong, and you should land at about ~1kHz or so
Anonymous at Sun, 26 May 2024 03:31:39 UTC No. 16193538
>>16192528
Yeah assume semi periodic conditions and then abuse fourier techniques as many times as a I can. Forcing interpolated smoothness I could probably do it a lot.
Anonymous at Sun, 26 May 2024 03:46:05 UTC No. 16193548
Not fast enough but best in terms of compression currently afaik
https://bellard.org/ts_zip/
Anonymous at Sun, 26 May 2024 03:52:58 UTC No. 16193557
>>16193376
kek
Anonymous at Sun, 26 May 2024 05:17:49 UTC No. 16193619
>>16192688
why is everyone so mean here
Anonymous at Sun, 26 May 2024 07:29:13 UTC No. 16193700
>>16192538
>Neuralink Compression Challeng
I suppose the want to do something bigger, where the amount of data would be overwhelming.
Anonymous at Sun, 26 May 2024 13:53:24 UTC No. 16194162
/sci/ is full of midwits
Anonymous at Sun, 26 May 2024 15:52:30 UTC No. 16194289
>>16194162
Don't need to announce yourself
Anonymous at Sun, 26 May 2024 16:52:51 UTC No. 16194365
>>16194289
cope
Anonymous at Sun, 26 May 2024 17:15:00 UTC No. 16194396
>>16193619
No repercussions and its part of the culture
Anonymous at Sun, 26 May 2024 19:59:55 UTC No. 16194633
>>16194365
Cut deep lmfao
Anonymous at Sun, 26 May 2024 23:28:54 UTC No. 16194954
>>16192897
Prize is for you to realize what sort of accomplishment this means when you submit your work.
Your small communist brain crying about billionaires isn't it.
Anonymous at Sun, 26 May 2024 23:31:34 UTC No. 16194959
>>16194954
>Wants to be paid
>You're a commie!
What did she mean by this?
Anonymous at Sun, 26 May 2024 23:33:22 UTC No. 16194962
>>16194959
>Resorting to class warfare talking points
>i'm not a commie
You sure do sound like one. And your small brain doesn't seem to understand the implication of submitting such a work to begin with nor the implication of the effect thereafter
Anonymous at Mon, 27 May 2024 06:29:52 UTC No. 16195359
Derpy Derp... Human brains don't use lossless compression. They mostly operate at the speed of sound. So how do eyes see faster than sound?
How does the brain deal with massive bandwidth streams like audio, video, and a hundred thousand simultaneously active nerve signals?
EMOJIS of more specifically, NEURAL GLYPHS.
A COMPRESSED SHORTHAND VERSION OF THE ORIGINAL CONCEPT.
You're a child in a car in motion. You look out the window, bored. A cow is in the field eating grass.
What does your brain see?
The motion senses know you're moving at high velocity, your sensory touch nerves report no injury or pain. What bandwidth is necessary to report this? Almost none, like hedonistic plateaus, once you reach a plateau of comfort, your nerves only report greater pleasure or greater pain. Therein the bandwidth is spent on variations from the current baseline.
You see the cow out the window. Your reptile brain screams, "Is that a threat? Can I eat it, fuck it, or will it inflict on me harm?" The ears report, "Cow is far away", Audio Snippet Glyph may be a quiet weak boring "moo". The eyes see a cow behind a sturdy wooden fence, no aggression noted, cow isn't noticing the car you're in. This is represented from the eyes to the result of the brain as a tiny cartoon Glyph (like a child might draw on paper in crayon for their parents fridge or an emoji icon of "bored cow"). The rest of the brain isn't getting every feature of the cow, cow color or gender or age, just the Glyph.
Bandwidth is almost nothing on that intra-brain chatter. Most of the brain doesn't really require a full audio signal or a full video replay or tactile memory of touching cow hide. A Glyph or Emoji is more than needed.
Think of your bandwidth. Audio is basically a 1D signal plus amplitudes and length. Touch is 3D plus amplitudes and a multiplier. Muscles are basically audio signal replays.
Video is a series of layers of antisignals. Since comes color bandwidth is designed to negate on a curve.
Anonymous at Mon, 27 May 2024 06:42:01 UTC No. 16195370
>>16195359
wat
Anonymous at Mon, 27 May 2024 06:57:51 UTC No. 16195383
>>16195359
>Human brains don't use lossless compression
Yes we do. Our memory function itself is lossy by design. Its designed to take in only relevant information and only for relevant time frame. Its completely not only lossy on encoding but lossy on decoding.
Anonymous at Mon, 27 May 2024 08:45:02 UTC No. 16195477
>>16195383
>Yes we do. Our memory function itself is lossy by design. Its designed to take in only relevant information and only for relevant time frame. Its completely not only lossy on encoding but lossy on decoding.
Yup!
If you're learning mathematical skills and that's important as a skillset for you, then learn to cycle through emotions as you learn.
The reason? YOUR EMOTIONAL STATES ARE TIED TO YOUR LEARNING RECALL COMPREHENSION. It's another form of compression. If you learned to do Algebra when you were irritated and bored, your memory will only accurately recall that information when irritated and bored. The normies will say, "that doesn't make sense". Everyone else will realize this is absolutely true.
Why do they stress you out, exhaust you and keep you distracted in Boot Camp? Because you will need these skills when you are tired, exhausted, confused,on a battlefield. They teach you tedious skills to get you to recall these skills when bored.
The skills you know when being chased by hungry bear in the woods aren't much useful when you're not being chased. Not all recall is locked to emotional states, but most of it is. Only sociopaths or Asperger's types aren't strongly linked to emotional recall as they're not strongly emotional. SKILL UNLOCK -- if you're depressed, force yourself to gain useful skills that would be beneficial to you when you are depressed in the future. If you over-leverage your skillsets to HAPPY TIME states, you will only be practically useful when happy. But you're not going to be happy all your life, leverage skills for when you're moody or depressed or bored or angry or horny. Be aware of your emotional skillset recall and max out your valuable learning times to be useful when your negative emotions are in overgear.
Anonymous at Mon, 27 May 2024 14:27:14 UTC No. 16195858
>>16192528
Just .ogg compresion is probably enough.
Anonymous at Mon, 27 May 2024 14:57:53 UTC No. 16195917
>>16192528
>Neuralink
Elon Musk? I will never again help that idiot in any way, at all.
Anonymous at Mon, 27 May 2024 15:15:21 UTC No. 16195952
>>16195917
By inventing smooth compresion, you don't help him, but his poor victims to communicate with exoskeleton.
Anonymous at Tue, 28 May 2024 06:27:43 UTC No. 16197197
>>16195952
kek
Anonymous at Wed, 29 May 2024 02:00:19 UTC No. 16198531
i won't let this thread die
Anonymous at Wed, 29 May 2024 04:43:07 UTC No. 16198665
>>16192826
Having constant open wounds in the skin is bad if you want to prevent infection. Doubly so on the scalp, where hair (almost ideal for catching and holding environmental contaminants) tends to grow. Look at how hard it is to maintain insulin pumps, for example.
Anonymous at Wed, 29 May 2024 23:00:51 UTC No. 16199847
>>16195508
Probably correlative pattern with neural wavefront propagation.
Anonymous at Wed, 29 May 2024 23:29:18 UTC No. 16199903
>>16195917
You probably have delusion of grandeur.
Anonymous at Thu, 30 May 2024 01:39:55 UTC No. 16200076
>>16192528
Too much noise in the data, first fix that shit otherwise is nothing but a huge waste of time to go lossless.
Anonymous at Thu, 30 May 2024 01:45:55 UTC No. 16200089
>>16192898
I prefer overwhelming power to exterminate anyone who disagrees with me.
Anonymous at Thu, 30 May 2024 07:32:26 UTC No. 16200419
>>16192826
No one wants a wired brainchip where you have to drag a long thick wire all around the house, outside the house, etc.
Anonymous at Thu, 30 May 2024 07:57:29 UTC No. 16200442
>>16200419
learn to read retard