fully open sourced neural networking operating via distributed computing for e.g. NLP (Natural Language Processing)
this is more a ruff idea i wanted to drop to get feedback. I don't have the expertise to do this kinda stuff myself.
The military complex is dumping millions and millions into NLP under the pretext of fighting terrorism. While I agree on the premise of fighting terrorism, I think the black-box approach of monitoring and analyzing human communication has to many inherent dangers of abuse to it.
Therefor I want to create the possibility of an alternative that is equal or even surpassing in effectiveness.
To achieve the needed computational power distributed computing comes to mind. This approach/solution should be achievable with just a fraction of the financial power needed for traditional neural networking systems. Volunteers could opt-in all over the world.
However, this approach comes with different obstacles. "How do you build a neural network that reliably can operate via distributed computing?"
One problem, that immediately comes to mind, is the exchange of calculations between different layers of the neural network and how the system as a whole has to react, if one of the connected devices disconnects before finishing it's layer's computations. (this disconnection problem is inherent to distributed computing)
I would suggest that building a neural network capable of operating via distributed computing is the main problem that needs to be tackled. Implementing specific uses like NLP should be a comparable small problem afterwards.
I thought the scene surely has some bright minds interested in solving problems like these :)
-- ima noob myself. my assumptions, or even the whole picture i have, may be wrong.
any input is welcome. (e.g. where else i could drop this idea to increase it's reach - if you know of some place, feel free to do so yourself.)
so far,
neca
PS: for reference, www.ibm.com
The military complex is dumping millions and millions into NLP under the pretext of fighting terrorism. While I agree on the premise of fighting terrorism, I think the black-box approach of monitoring and analyzing human communication has to many inherent dangers of abuse to it.
Therefor I want to create the possibility of an alternative that is equal or even surpassing in effectiveness.
To achieve the needed computational power distributed computing comes to mind. This approach/solution should be achievable with just a fraction of the financial power needed for traditional neural networking systems. Volunteers could opt-in all over the world.
However, this approach comes with different obstacles. "How do you build a neural network that reliably can operate via distributed computing?"
One problem, that immediately comes to mind, is the exchange of calculations between different layers of the neural network and how the system as a whole has to react, if one of the connected devices disconnects before finishing it's layer's computations. (this disconnection problem is inherent to distributed computing)
I would suggest that building a neural network capable of operating via distributed computing is the main problem that needs to be tackled. Implementing specific uses like NLP should be a comparable small problem afterwards.
I thought the scene surely has some bright minds interested in solving problems like these :)
-- ima noob myself. my assumptions, or even the whole picture i have, may be wrong.
any input is welcome. (e.g. where else i could drop this idea to increase it's reach - if you know of some place, feel free to do so yourself.)
so far,
neca
PS: for reference, www.ibm.com