4 ms·

Neurokernel: Emulating the Fruit Fly Brain

46 points by T-A 12y ago

adrianN 12y ago
There is a similar project called OpenWorm, that aims to simulate a nematode completely. It's pretty cool, you should check it out http://www.openworm.org/ http://www.openworm.org/
guru_meditation 12y ago
Quoting from the course front page: Students with extensive software engineering experience (systems software, parallel programming or computer graphics) are strongly encouraged to apply. Almost makes me want to go back to grad school. Look at these folks, they made GPU-enabled code available as Python's SciKit[1] But then I see they're patenting stuff left and right [2] and my enthusiasm for this project dwindles.... [1] http://www.bionet.ee.columbia.edu/code/scikits.cuda http://www.bionet.ee.columbia.edu/code/scikits.cuda [2] http://www.bionet.ee.columbia.edu/patents/ http://www.bionet.ee.columbia.edu/patents/
petegrif 12y ago
There is another project in prospect - CruzPa - that attempts to model a human brain specifically selected for its apparent simplicity. http://www.buzzfeed.com/ilanbenmeir/the-most-controversial-things-ted-cruzs-dad-has-ever-said#.rd5ajVB8G http://www.buzzfeed.com/ilanbenmeir/the-most-controversial-t...
comrade1 12y ago
Based on the current rate of processor development we will probably be able to model a human brain in realtime around 2045 on a supercomputer. We can model the processing of a human now but it is exceedingly slow, and it is just toy calculations. So even if we can model a mind in 2045 it will may be a long time after that before it can be done in a meaningful way. In the meantime we will probably be able to model humans in a more simplistic way. Our external interactions are simple. If we can record all interactions of a person over time we can develop a cognitive profile and develop a 'beta' copy of that person. Not a real thinking AI, but to someone interacting with that copy essentially the real thing. People are working on this topic now but they're a long way away. I'm guessing 2035.
- Retric 12y ago
  We have trouble accurately modeling how a single protean folds. A cell is vastly more complex. And a brain is far more so. So, at best we are talking about some model of a human brain at some detail level. IMO, a timescale on it is silly when we don't know how complex the model needs to be to be useful.
  - comrade1 12y ago
    They're able to model about a second of brain processing in about an hour now on our largest computers. The challenge of course is the parallel nature of our processing. Based on current rates of processing growth they can probably model a brain in realtime in 2045. http://www.telegraph.co.uk/technology/10567942/Supercomputer-models-one-second-of-human-brain-activity.html http://www.telegraph.co.uk/technology/10567942/Supercomputer...
    - Retric 12y ago
      "with a single second’s worth of activity from just one per cent" So, at some accuracy level they modeled 1% of a brain. However, while 'the most accurate yet' yet they did not model things like cell death and as ~10,000 neurons die a day it's still an important process even in adult brains. With children it's extremely important as over 1/2 of the brain cells your born with die by 18. Granted on a second by second basis cell death is not that important, but if you want to model a human brain vs say an abstract neural net then accuracy is important. Sure, other things like cellular energy expenditure, hormones, orientation, gravity, blood flow, oxygen levels, immune response, might not seem important. But, again do you want to model a brain or make a fun simulation.
      - comrade1 12y ago
        That's where the 2045 date comes from. That article mentions the 1% issue, and based on that someone came up with the date 2040/2050.
noiv 12y ago
According to the specs a fruit fly is smarter than my mobile.

46 points by T-A 12y ago

There is a similar project called OpenWorm, that aims to simulate a nematode completely. It's pretty cool, you should check it out http://www.openworm.org/ http://www.openworm.org/

guru_meditation 12y ago

Quoting from the course front page: Students with extensive software engineering experience (systems software, parallel programming or computer graphics) are strongly encouraged to apply. Almost makes me want to go back to grad school. Look at these folks, they made GPU-enabled code available as Python's SciKit[1] But then I see they're patenting stuff left and right [2] and my enthusiasm for this project dwindles.... [1] http://www.bionet.ee.columbia.edu/code/scikits.cuda http://www.bionet.ee.columbia.edu/code/scikits.cuda [2] http://www.bionet.ee.columbia.edu/patents/ http://www.bionet.ee.columbia.edu/patents/

petegrif 12y ago

There is another project in prospect - CruzPa - that attempts to model a human brain specifically selected for its apparent simplicity. http://www.buzzfeed.com/ilanbenmeir/the-most-controversial-things-ted-cruzs-dad-has-ever-said#.rd5ajVB8G http://www.buzzfeed.com/ilanbenmeir/the-most-controversial-t...

comrade1 12y ago

Based on the current rate of processor development we will probably be able to model a human brain in realtime around 2045 on a supercomputer. We can model the processing of a human now but it is exceedingly slow, and it is just toy calculations. So even if we can model a mind in 2045 it will may be a long time after that before it can be done in a meaningful way. In the meantime we will probably be able to model humans in a more simplistic way. Our external interactions are simple. If we can record all interactions of a person over time we can develop a cognitive profile and develop a 'beta' copy of that person. Not a real thinking AI, but to someone interacting with that copy essentially the real thing. People are working on this topic now but they're a long way away. I'm guessing 2035.

Retric 12y ago
We have trouble accurately modeling how a single protean folds. A cell is vastly more complex. And a brain is far more so. So, at best we are talking about some model of a human brain at some detail level. IMO, a timescale on it is silly when we don't know how complex the model needs to be to be useful.
- comrade1 12y ago
  They're able to model about a second of brain processing in about an hour now on our largest computers. The challenge of course is the parallel nature of our processing. Based on current rates of processing growth they can probably model a brain in realtime in 2045. http://www.telegraph.co.uk/technology/10567942/Supercomputer-models-one-second-of-human-brain-activity.html http://www.telegraph.co.uk/technology/10567942/Supercomputer...
  - Retric 12y ago
    "with a single second’s worth of activity from just one per cent" So, at some accuracy level they modeled 1% of a brain. However, while 'the most accurate yet' yet they did not model things like cell death and as ~10,000 neurons die a day it's still an important process even in adult brains. With children it's extremely important as over 1/2 of the brain cells your born with die by 18. Granted on a second by second basis cell death is not that important, but if you want to model a human brain vs say an abstract neural net then accuracy is important. Sure, other things like cellular energy expenditure, hormones, orientation, gravity, blood flow, oxygen levels, immune response, might not seem important. But, again do you want to model a brain or make a fun simulation.
    - comrade1 12y ago
      That's where the 2045 date comes from. That article mentions the 1% issue, and based on that someone came up with the date 2040/2050.

noiv 12y ago

According to the specs a fruit fly is smarter than my mobile.