5 ms·

My personal theory on this is twofold: 1.) High achievers are used to pushing at things to win. Creativity is like a garden - you have to do some tending, bu

My personal theory on this is twofold: 1.) High achievers are used to pushing at things to win. Creativity is like a garden - you have to do some tending, but then you have to back off and let the plants grow. Fiddling doesn't help, in fact you're liable to end up killing the plants. 2.) Smart people actually learn too fast. Creativity requires very broad neural connectivity, and I think fast learning tends to produce neural networks with sparse connectivity to different areas. This is supported to a degree by learning in artificial neural networks. When the learning rate in artificial neural networks is too fast, this can cause the network parameters to converge prematurely. This premature convergence typically results in poor generalization performance. It is also worth noting that human brains mature more slowly than those of previous hominids and great apes.

nostrebored 10y ago
Using neural networks as a machine learning concept to explain actual concepts in Neuroscience is a huge overstep. People have always thought the most advanced technology can be analogous to cognition, see steam engines and clocks.
- OisinMoran 10y ago
  Your second sentence is definitely true and very interesting but maybe you'd agree that the gap between our most advanced technology and the brain is decreasing to the point where this will eventually be right.
  - princeb 10y ago
    > gap between our most advanced technology and the brain is decreasing first we need to understand the brain.
- psyc 10y ago
  And those were progressively better analogies. These are great terms to think in, and it's not an overstep. It doesn't have to be absolutely true to be very useful for understanding.
- CuriouslyC 10y ago
  Artificial neural networks are much closer to neurons and brains than economic models are to markets, but we give economists a pass. Perhaps we should stop listening to them and revoke their publishing privileges? Anyhow, the point wasn't to say Brains == ANN, rather it was to show how learning too quickly has consequences in other models of learning. This was mostly so people wouldn't just dismiss the idea out of hand as having no justification (which people seem to want to do regardless).
  - nostrebored 10y ago
    Really? I think that we have a much better understanding of markets than Neuroscience. The human brain is an immensely complex system. We just figured out that lungs may produce the majority of our blood cells -- do you really have so much faith in a field that routinely misrepresents results to fit their hypotheses?
Retric 10y ago
2) This falls into the just world fallacy. Intelligence does not seem to come with real drawbacks. Highly intelligent people often have upper end reflexes, are unusual attractive, highly creative, social, and or content. They are more often taller and live longer as well. The trappings of intelligence are different, but very high raw intelligence is seen across most walks of life including actors, athletes, and salesmen.
- ziikutv 10y ago
  I would also like to point out that while there might not be any direct drawbacks, think about others. How many kids put down those smarter to feel better about themselves? Another disagreement: Not all intelligent people are highly social. Many are introverted. Intelligent or introverted kids get bullied the most.
  - Retric 10y ago
    Your assuming kids are accurately measuring intelegence and then picking on the smartest person. Instead they are picking on an outlier who may be less intelegent than other people in the population with different interests.
    - ziikutv 10y ago
      Fair point.
- j9461701 10y ago
  This is only true up to a point. Past ~125 IQ, you start running into serious social problems that mean such individuals are much less likely to obtain leadership positions, join intellectually challenging professions, or obtain advanced degrees. https://polymatharchives.blogspot.ca/2015/01/the-inappropriately-excluded.html https://polymatharchives.blogspot.ca/2015/01/the-inappropria... http://psycnet.apa.org/psycinfo/2017-14279-001/ http://psycnet.apa.org/psycinfo/2017-14279-001/
  - 10y ago
deepnotderp 10y ago
Repeat after me: Artificial Neural Networks/DL =! Biological Neural Networks
- CuriouslyC 10y ago
  You'd have to be pretty ignorant to assume they are exactly analogous, but to assume that there is no transfer is equally ignorant. Artificial neural networks are much closer to neurons and brains than economic models are to markets, yet we bet the farm on forecasting (with pretty bad results, I might add). The repeat after me meme is really condescending by the way, might want to avoid using that.
  - deepnotderp 10y ago
    What are the analogous components? Backpropagation is biologically implausible.
    - CuriouslyC 10y ago
      Dopamine released from task success primes long term potentiation, effectively acting as the analog of the cost function's gradient. The larger the dopamine hit (generally) the bigger the learning step. You are correct that neurons can't do backprop. Keep in mind that the networks of the brain aren't straight feedforward, they're recurrent. In order to provide temporal control of activation propagation throughout the network, inhibitory neurons are needed. Thus, instead of "tweaking weights" backwards in the network, the brain learns to activate inhibitory neurons to provide forward feedback against activation.The mechanism is different but the learning effect is pretty similar.

nostrebored 10y ago

Using neural networks as a machine learning concept to explain actual concepts in Neuroscience is a huge overstep. People have always thought the most advanced technology can be analogous to cognition, see steam engines and clocks.

OisinMoran 10y ago
Your second sentence is definitely true and very interesting but maybe you'd agree that the gap between our most advanced technology and the brain is decreasing to the point where this will eventually be right.
- princeb 10y ago
  > gap between our most advanced technology and the brain is decreasing first we need to understand the brain.
psyc 10y ago
And those were progressively better analogies. These are great terms to think in, and it's not an overstep. It doesn't have to be absolutely true to be very useful for understanding.
CuriouslyC 10y ago
Artificial neural networks are much closer to neurons and brains than economic models are to markets, but we give economists a pass. Perhaps we should stop listening to them and revoke their publishing privileges? Anyhow, the point wasn't to say Brains == ANN, rather it was to show how learning too quickly has consequences in other models of learning. This was mostly so people wouldn't just dismiss the idea out of hand as having no justification (which people seem to want to do regardless).
- nostrebored 10y ago
  Really? I think that we have a much better understanding of markets than Neuroscience. The human brain is an immensely complex system. We just figured out that lungs may produce the majority of our blood cells -- do you really have so much faith in a field that routinely misrepresents results to fit their hypotheses?

Retric 10y ago

2) This falls into the just world fallacy. Intelligence does not seem to come with real drawbacks. Highly intelligent people often have upper end reflexes, are unusual attractive, highly creative, social, and or content. They are more often taller and live longer as well. The trappings of intelligence are different, but very high raw intelligence is seen across most walks of life including actors, athletes, and salesmen.

ziikutv 10y ago
I would also like to point out that while there might not be any direct drawbacks, think about others. How many kids put down those smarter to feel better about themselves? Another disagreement: Not all intelligent people are highly social. Many are introverted. Intelligent or introverted kids get bullied the most.
- Retric 10y ago
  Your assuming kids are accurately measuring intelegence and then picking on the smartest person. Instead they are picking on an outlier who may be less intelegent than other people in the population with different interests.
  - ziikutv 10y ago
    Fair point.
j9461701 10y ago
This is only true up to a point. Past ~125 IQ, you start running into serious social problems that mean such individuals are much less likely to obtain leadership positions, join intellectually challenging professions, or obtain advanced degrees. https://polymatharchives.blogspot.ca/2015/01/the-inappropriately-excluded.html https://polymatharchives.blogspot.ca/2015/01/the-inappropria... http://psycnet.apa.org/psycinfo/2017-14279-001/ http://psycnet.apa.org/psycinfo/2017-14279-001/
- 10y ago

deepnotderp 10y ago

Repeat after me: Artificial Neural Networks/DL =! Biological Neural Networks

CuriouslyC 10y ago
You'd have to be pretty ignorant to assume they are exactly analogous, but to assume that there is no transfer is equally ignorant. Artificial neural networks are much closer to neurons and brains than economic models are to markets, yet we bet the farm on forecasting (with pretty bad results, I might add). The repeat after me meme is really condescending by the way, might want to avoid using that.
- deepnotderp 10y ago
  What are the analogous components? Backpropagation is biologically implausible.
  - CuriouslyC 10y ago
    Dopamine released from task success primes long term potentiation, effectively acting as the analog of the cost function's gradient. The larger the dopamine hit (generally) the bigger the learning step. You are correct that neurons can't do backprop. Keep in mind that the networks of the brain aren't straight feedforward, they're recurrent. In order to provide temporal control of activation propagation throughout the network, inhibitory neurons are needed. Thus, instead of "tweaking weights" backwards in the network, the brain learns to activate inhibitory neurons to provide forward feedback against activation.The mechanism is different but the learning effect is pretty similar.