neural networking via distributed computing
category: general [glöplog]
huhu,
i've created my first wanted request ever, and wanted to create a discussion about the topic the project tackles and/or the project itself.
even the most simple input like "good/bad idea" or "u stupid! neva gon do" would be welcome. :)
so far
i've created my first wanted request ever, and wanted to create a discussion about the topic the project tackles and/or the project itself.
even the most simple input like "good/bad idea" or "u stupid! neva gon do" would be welcome. :)
so far
Maybe check up on the state of the art first:
https://research.googleblog.com/2017/04/federated-learning-collaborative.html
https://research.googleblog.com/2017/04/federated-learning-collaborative.html
What are you planning that needs so much power? I mean we fairly trivially do NLP on mobile these days...
Skynet!!
psonice: !!
we hope you make yourself at home here in our lovely mobile app, we are totes legit and we're sure you'll enjoy this smashing game, which features lots of cars and lots of super chases
we hope you make yourself at home here in our lovely mobile app, we are totes legit and we're sure you'll enjoy this smashing game, which features lots of cars and lots of super chases
Fell:
Advertising: 97%
Suspicious: 84%
Car stunt driver: 4%
Advertising: 97%
Suspicious: 84%
Car stunt driver: 4%
Sesse:
thanks for the link. i am atm reading up (right now somewhere in one of the linked papers...much text) but there doesn't seem to be source code of the keyboard available.
their project isn't open source, or was i just to stupid to find it? reading the actual source would help me understand what they are doing way more than reading abstract stuff.
as far as i can tell by now: they do ...some... calculations and encrypt them before sending to their cloud server. this cloud server seems to do big chunks of calculations - not only a data hub.
i can't tell, if they actually solved the problem of a dynamic solution (which i had in mind), yet. Could be, that every end-device is just doing the exact same layer calculations and send those to the cloud servers for further neural network layers. so as to read: a work around to the privacy laws to collect massive chunks of data (plz correct me, if i am wrong... sry that i like to speculate with just minor understanding of the topic ;p)
since this thread seems to have developed into mainly a discussion of the topic at hand and not the project (:D), may i ask you what you think of black-box solutions monitoring/analyzing our communication?
i have a strongly opinionated stance. to put it short: I'm not ok with it.
psonice:
actually... my main goal is to develop an open source solution so everyone (in this case every community, since distributed computing..), could decide themselves what they use it for. (like *unix/windows; firefox/IE [...])
i have some ideas myself, but convincing others to opt-in to the distributed computing is a whole nother issue (google just put it in without much end-user choice... that is an easy and dirty way to do it...)
e.g. cross-checking some claims. (are they using their black-box for what they are claiming? - do the same thing and if you find a solution way quicker, they prolly don't or are just stupid...)
ah, well... should dissect my virtual keyboard now. looking what actually is going on.
(don't even know where to start, yet... like said: me noob xD)
thanks for the link. i am atm reading up (right now somewhere in one of the linked papers...much text) but there doesn't seem to be source code of the keyboard available.
their project isn't open source, or was i just to stupid to find it? reading the actual source would help me understand what they are doing way more than reading abstract stuff.
as far as i can tell by now: they do ...some... calculations and encrypt them before sending to their cloud server. this cloud server seems to do big chunks of calculations - not only a data hub.
i can't tell, if they actually solved the problem of a dynamic solution (which i had in mind), yet. Could be, that every end-device is just doing the exact same layer calculations and send those to the cloud servers for further neural network layers. so as to read: a work around to the privacy laws to collect massive chunks of data (plz correct me, if i am wrong... sry that i like to speculate with just minor understanding of the topic ;p)
since this thread seems to have developed into mainly a discussion of the topic at hand and not the project (:D), may i ask you what you think of black-box solutions monitoring/analyzing our communication?
i have a strongly opinionated stance. to put it short: I'm not ok with it.
psonice:
Quote:
What are you planning that needs so much power?
actually... my main goal is to develop an open source solution so everyone (in this case every community, since distributed computing..), could decide themselves what they use it for. (like *unix/windows; firefox/IE [...])
i have some ideas myself, but convincing others to opt-in to the distributed computing is a whole nother issue (google just put it in without much end-user choice... that is an easy and dirty way to do it...)
e.g. cross-checking some claims. (are they using their black-box for what they are claiming? - do the same thing and if you find a solution way quicker, they prolly don't or are just stupid...)
ah, well... should dissect my virtual keyboard now. looking what actually is going on.
(don't even know where to start, yet... like said: me noob xD)
Gboard isn't open source, no.
mhm... maybe train a version of tensorflow to find functions in code, that are functionally identically - - afterwards let it run over Gboard and check against itself, to find the stuff that's not copy&paste, which should be the distributed computing stuff.
could that work? :D or too big of a project itself - - better to simply manually look over it?
could that work? :D or too big of a project itself - - better to simply manually look over it?
@psonice ohhhh I was reading as neuro-linguistic programming :P
necaremus: Have you already considered to go make a demo about it?
@necaremus: It won't work. Finding out if two general functions are functionally identical is an undecidable problem (it would give you a solution to the halting problem, for one).
havoc: not until now. interesting idea :)
Sesse: mhm... now that you mention it, it sounds logical. my mind instantly got busy trying to find a work around, thou :D approximation maybe?
____
does anyone here has experience with tensorflow and/or gensim?
i am undecided on which library would be more fitting for the task. gensim seems to already have some support for distribution (i just started to read up), while google uses tensorflow.
Sesse: mhm... now that you mention it, it sounds logical. my mind instantly got busy trying to find a work around, thou :D approximation maybe?
____
does anyone here has experience with tensorflow and/or gensim?
i am undecided on which library would be more fitting for the task. gensim seems to already have some support for distribution (i just started to read up), while google uses tensorflow.