Well, training is basically the process of getting neural activity into some new local equilibrium from an old one. That's broadly how back-propagation and Kohonen-style networks work
That's how *any* dynamical system works by definition (assuming they have a stable state)
But that doesn't make it learning, and it bears no resemblence to the fundamental algorithm of back prop.
Iit's just a mass of cells that happens to have a way to communicate with each other that's convenient for the application.
Actually it's horrifically inconvenient. The behavior to stabilize a plane like this could probably be approximated with a two variable dynamic equation.
Instead they've got pipes and aerators, silicon embedded growth matrices, at least 2 computers, amplifiers, stimulators...
Yea, I think it's a stretch to call it convenient.
Re:One question...
on
Flying By Brain
·
· Score: 5, Interesting
Someone correct me if I'm wrong, but it's hard to tell because these guys specifically avoid using the technical terminology of the LTP (Long Term Potentiation) literature, probably because they know they aren't getting it and don't want to step into that minefield.
As near as I can tell from their paper at:
http://www.neuro.gatech.edu/groups/potter/papers/D agstuhlAIBakkumpreprint.pdf
the network is not "learning". Rather, they are setting up the system so that the inherent properties of the neurons cause the correct response to the feedback it receives from the environment.
The real knowledge about the task is built into the systems that interface with the neurons.
As an analogy, the neuron is behaving like a spring in a mechanical system, it has some basic fundamental properties that are statistically predictable, and the system around the spring expects it to behave thusly. But because it's a complex system it may take time for the system to settle into the stable state, hence it looks as if the network "learns", when really it's a system of springs settling into an equilibrium.
Not to understate their technical accomplishments. They've done amazing things with cultured neurons. But this is not about reward and punishment, the network is far too simple for such words to have any meaning. It may not even be about learning in the sense of permanently modifying synaptic connections. I can't tell from my first read through, and that's what really sets off the alarm bells.
They also avoid the obvious experiment that should be done if they think long term plasticity is involved. (ie, can it still navigate the next day?)
Science is not a pure discipline, and is greatly affected by political influence, exerted through funding schemes.
Right now, Global warming is the popular opinion, and because of this, there's a huge bias towards pro-global warming research. You can get away with publishing awfully specious results as long as they say the right thing. But publish an anti-Warming article, and it will have to pass far more stringent test.
http://cdiac.esd.ornl.gov/trends/co2/graphics/vo st ok.co2.gif
Which are extrapolated atmospheric CO2 measurements from the ice core in Antarctica, and then tack on the modern CO2 measurements from the actual atmosphere in a different climate.
That they don't even mention the caveats of this dangerously speculative merging of incompatible data sets makes it clear that they are more worried about convincing you than finding the truth.
The two methodologies (trapped CO2 measurement in ice vs CO2 measurement in the atmosphere) cannot be merged like this, even discounting the vast differences in climate between Hawaii and Antarctica.
For failing to mention glaring caveats that should be obvious to anyone with a critical scientific eye, I have to dub this alarmist propaganda which is informative and interesting, but not empirical.
That's an excellent point. But the same applies to the same money being applied to the environment, so it's not really a point we can worry about. Corruption just happens.
And at any point in time, the respective existing model is all you've got to base a sound argument on
Absolutely not true. You've also got historical data, which in this case indicate that the earth has repeatedly been through cycles of variation far more extreme than this one that mankind had no role in.
We're so pre-occupied with the next 10 years in all of these proposals, but realistically it's the next several decades/century that will really determine how the environment play out.
Kyoto is ignoring the real long term solution (elevate the third world to a well-fed and stable system that starts worrying about the environment instead of tomorrow's breakfast) in order to apply a feel good band-aide patch that will let us pat ourselves on the back.
Then when the real problems occur in 50 years in spite of such nearsighted efforts, we'll all be screaming "don't blame us, we implemented Kyoto, it's all the deforestation by slash& burn farming done by poor farmers in Brazil!"
I'm against Kyoto, but not because I don't care about the environment. Quite the contrary.
The earth's carbon sinks are not static in capacity. Everything is interlocked feedback cycles. As CO2 goes up, so too does the growth rate of all vegetation.
It is the naive simplicity of the mathematics used by many lay-men(and sometimes experts) in their discussions of climate change that cause me to seriously doubt their prediction.
Check out this web page for example http://www.hydrogen.co.uk/h2_now/journal/ articles/ 2_global_warming.htm
which tries to use *addition* to predict changes in CO2. We produce X billion tons, the amazon absorbs Y billion tons, net change is X-Y billion tons.
This approach is as hopelessly naive as trying to calculate the flight dynamics of the space shuttle with natural numbers.
That's just not how it works in a real dynamic system and alarmist crap like this only serves to push through ridiculous laws like Kyoto, the funding for which could bring food and water to a huge proportion of the third world instead of affecting some laughable 7% of the annual *human* CO2 output.
Get those people fed and industrialized, and they'll stop cutting down their own forests, start going to school, and add their share of brainpower to the world's thinktank.
Boston Bike Friendly?! I biked 6 miles every day for years in Cambridge & Brookline, and it's no picnic. Downtown Boston I've biked through many times and it's even worse.
Small winding streets give you insufficient room to avoid being doored and limit visibility. It's playing russian roulette every day. I've been lucky so far *knocks wood*.
Heavy traffic load makes motorists impatient, agitated and unpredictable. Busses swerve into and out of what few "bike lanes" exist.
Social problems and gaming streaks are a bit of a chicken and egg problem. It's not always clear which caused which.
A minor setback in the social agenda can cause a spree of gaming, which removes one from the loop further, making it ever harder to get back into a pattern of interacting with friends.
It's really a downward spiral which, for most people, has a rebound at the end.
I've gotten used to these cycles, and as I age they grow fewer and farther between. Now when I see a game coming that I'll expect to play, I warn my friends ahead of time. I spend a few weeks in a cocoon, play the hell out of it and don't feel guilty, knowing that it'll be gone soon and I'll be back to normal.
This kind of play hard/cold turkey cycle works very well for me, but it's difficult to learn and only works with games that don't require serious long term committments (not EQ).
What you are proposing is the spontaneous creation of energy. Give something a gentle push out the airlock, wait a few years, and the next time you encounter that object, it'll be a catastrophic impact?
Sorry, doesn't work. Relative velocity of the two objects will stay relatively similar unless an outside force interacts.
There are more hurricanes this year: it's because it's getting warmer!
I have a hard time putting much stock in climatologists claiming to understand dynamic systems. The problem is so hideously complicated, neglecting one feedback loop can throw all of the predictions out of whack.
Slashdot Mirror
Well, training is basically the process of getting neural activity into some new local equilibrium from an old one. That's broadly how back-propagation and Kohonen-style networks work
That's how *any* dynamical system works by definition (assuming they have a stable state)
But that doesn't make it learning, and it bears no resemblence to the fundamental algorithm of back prop.
Apparently they had that technique done a few years back, and yes it is very impressive. They claim cultures living up to a year!
Iit's just a mass of cells that happens to have a way to communicate with each other that's convenient for the application.
Actually it's horrifically inconvenient. The behavior to stabilize a plane like this could probably be approximated with a two variable dynamic equation.
Instead they've got pipes and aerators, silicon embedded growth matrices, at least 2 computers, amplifiers, stimulators...
Yea, I think it's a stretch to call it convenient.
Someone correct me if I'm wrong, but it's hard to tell because these guys specifically avoid using the technical terminology of the LTP (Long Term Potentiation) literature, probably because they know they aren't getting it and don't want to step into that minefield.
D agstuhlAIBakkumpreprint.pdf
As near as I can tell from their paper at:
http://www.neuro.gatech.edu/groups/potter/papers/
the network is not "learning". Rather, they are setting up the system so that the inherent properties of the neurons cause the correct response to the feedback it receives from the environment.
The real knowledge about the task is built into the systems that interface with the neurons.
As an analogy, the neuron is behaving like a spring in a mechanical system, it has some basic fundamental properties that are statistically predictable, and the system around the spring expects it to behave thusly. But because it's a complex system it may take time for the system to settle into the stable state, hence it looks as if the network "learns", when really it's a system of springs settling into an equilibrium.
Not to understate their technical accomplishments. They've done amazing things with cultured neurons. But this is not about reward and punishment, the network is far too simple for such words to have any meaning. It may not even be about learning in the sense of permanently modifying synaptic connections. I can't tell from my first read through, and that's what really sets off the alarm bells.
They also avoid the obvious experiment that should be done if they think long term plasticity is involved. (ie, can it still navigate the next day?)
Oh thank god. I didn't realize the motto had become Publish *and* Perish.
Store all my emails? um... no thanks? please?
That is a paper thin and naive argument.
There are all kinds of examples in which one person's values must be pushed on to others for the good of society.
You sound about 15. I'd like you to go back and reread this post of yours when you're 30.
You've been bitching all along that parents should be responsible for their kids, when realistically, it's impossible in many situations.
Now they're finally given a law to make it possible and you bitch about that too.
Parents are not superbeings. Cut them some slack, it's hard enough.
Of how easy our industrial centers are to spot from orbit.
We micronians are entirely ignorant of space-war tactics.
At the risk of losing your job for trumpeting the dominant scientific paradigm? right.
Also next time try not posting as an AC
Agreed.
Science is not a pure discipline, and is greatly affected by political influence, exerted through funding schemes.
Right now, Global warming is the popular opinion, and because of this, there's a huge bias towards pro-global warming research. You can get away with publishing awfully specious results as long as they say the right thing. But publish an anti-Warming article, and it will have to pass far more stringent test.
They take the data from here:
http://cdiac.esd.ornl.gov/trends/co2/graphics/v
Which are extrapolated atmospheric CO2 measurements from the ice core in Antarctica, and then tack on the modern CO2 measurements from the actual atmosphere in a different climate.
That they don't even mention the caveats of this dangerously speculative merging of incompatible data sets makes it clear that they are more worried about convincing you than finding the truth.
The two methodologies (trapped CO2 measurement in ice vs CO2 measurement in the atmosphere) cannot be merged like this, even discounting the vast differences in climate between Hawaii and Antarctica.
For failing to mention glaring caveats that should be obvious to anyone with a critical scientific eye, I have to dub this alarmist propaganda which is informative and interesting, but not empirical.
That's an excellent point. But the same applies to the same money being applied to the environment, so it's not really a point we can worry about. Corruption just happens.
And at any point in time, the respective existing model is all you've got to base a sound argument on
Absolutely not true. You've also got historical data, which in this case indicate that the earth has repeatedly been through cycles of variation far more extreme than this one that mankind had no role in.
There's an overwhelming amount of biased moderation going on in thread, more than is usual even for a global warming debate.
Can mods please stop shooting down well written and insightful arguments from the other side? What are you afraid of? being convinced?
It does in the long run.
We're so pre-occupied with the next 10 years in all of these proposals, but realistically it's the next several decades/century that will really determine how the environment play out.
Kyoto is ignoring the real long term solution (elevate the third world to a well-fed and stable system that starts worrying about the environment instead of tomorrow's breakfast) in order to apply a feel good band-aide patch that will let us pat ourselves on the back.
Then when the real problems occur in 50 years in spite of such nearsighted efforts, we'll all be screaming "don't blame us, we implemented Kyoto, it's all the deforestation by slash& burn farming done by poor farmers in Brazil!"
I'm against Kyoto, but not because I don't care about the environment. Quite the contrary.
The earth's carbon sinks are not static in capacity. Everything is interlocked feedback cycles. As CO2 goes up, so too does the growth rate of all vegetation.
/ articles/ 2_global_warming.htm
It is the naive simplicity of the mathematics used by many lay-men(and sometimes experts) in their discussions of climate change that cause me to seriously doubt their prediction.
Check out this web page for example
http://www.hydrogen.co.uk/h2_now/journal
which tries to use *addition* to predict changes in CO2. We produce X billion tons, the amazon absorbs Y billion tons, net change is X-Y billion tons.
This approach is as hopelessly naive as trying to calculate the flight dynamics of the space shuttle with natural numbers.
That's just not how it works in a real dynamic system and alarmist crap like this only serves to push through ridiculous laws like Kyoto, the funding for which could bring food and water to a huge proportion of the third world instead of affecting some laughable 7% of the annual *human* CO2 output.
Get those people fed and industrialized, and they'll stop cutting down their own forests, start going to school, and add their share of brainpower to the world's thinktank.
So then the NASA engineers mistakenly fix a problem that doesn't exist. Either way, bad news.
Well it does do that nice little turn at the beginning.
Although maybe that's a painstakingly pre-programmed maneuver, and not a reaction to approaching the edge of the table.
I've got $50 says this perfectly functional satellite is ruined because the AI mistakenly thinks there's a problem and tries to fix it.
KISS.
Boston Bike Friendly?! I biked 6 miles every day for years in Cambridge & Brookline, and it's no picnic. Downtown Boston I've biked through many times and it's even worse.
Small winding streets give you insufficient room to avoid being doored and limit visibility. It's playing russian roulette every day. I've been lucky so far *knocks wood*.
Heavy traffic load makes motorists impatient, agitated and unpredictable. Busses swerve into and out of what few "bike lanes" exist.
And the potholes...
You'll care if they start charging you for it.
Social problems and gaming streaks are a bit of a chicken and egg problem. It's not always clear which caused which.
A minor setback in the social agenda can cause a spree of gaming, which removes one from the loop further, making it ever harder to get back into a pattern of interacting with friends.
It's really a downward spiral which, for most people, has a rebound at the end.
I've gotten used to these cycles, and as I age they grow fewer and farther between. Now when I see a game coming that I'll expect to play, I warn my friends ahead of time. I spend a few weeks in a cocoon, play the hell out of it and don't feel guilty, knowing that it'll be gone soon and I'll be back to normal.
This kind of play hard/cold turkey cycle works very well for me, but it's difficult to learn and only works with games that don't require serious long term committments (not EQ).
Sanity checking reveals you are in error.
What you are proposing is the spontaneous creation of energy. Give something a gentle push out the airlock, wait a few years, and the next time you encounter that object, it'll be a catastrophic impact?
Sorry, doesn't work. Relative velocity of the two objects will stay relatively similar unless an outside force interacts.
It's getting warmer: we're causing it!
There are more hurricanes this year: it's because it's getting warmer!
I have a hard time putting much stock in climatologists claiming to understand dynamic systems. The problem is so hideously complicated, neglecting one feedback loop can throw all of the predictions out of whack.