Thread: some ideas about the brain

Main Ideas

Does this belong in the biology section or computer section? I have some evidence to support these idea, although they're mainly inferences/reasoning.
These ideas are mainly for a neural network designed by Jeff Hawkins, called Hierarchial-Temporal Memory.
Here are some major topics which I think are important to the creation of brain-based machines. I'll explain more details when I have time, but since it will probably take me a while to write everything I want to, here is something more general.
Neurotransmitters affect large groups of neurons at once, in a variety of ways. Neurotransmitters can adjust thresholds, affect different types of neurons differently, or change the amount of time a neuron collects input before it can fire again. Neurotransmitters are highly involved in emotions, which are fundamentally an instinct to act a certain way (this idea is not entirely my own. Based on the Lovheim cube of emotions.). That could be seen as giving an artificial neural network a hint as to what it is meant to do. I think neurotransmitters create a hierarchy of goals, which helps an animal achieve the next highest quality of living, similar to Maslow's hierarchy of needs. Neurotransmitters are also important to learning, especially during sleep. While a person is asleep, new ideas can be explored without any danger. During sleep, the neocortex recieves random data from the pons, which could be like looking at a sequence of scribbles, and interpretting them as dreams. This is more effective because of the neurotransmitters involved in sleep. Besides adjusting plasticity and how quickly the brain changes focus, the neurotransmitters involved in sleep cause the neocortex to pay much less attention to detail, which is useful for interpretting random information without being distracted by noise.
The life cycle is important to development. As embryos, we learn the most basics laws of the universe, such as it is 3-d. As babies, we learn about food and comfort, and then speech. As children, we learn complex social skills. As adults, we learn to pay taxes and have a job. Imagine if a baby were given a job. It would be clueless, and it wouldn't learn anything. Instead of giving babies jobs, we slowly increase the complexity of their lives until they reach adulthood. That idea could be applied to neural networks.
The rest of these ideas are underdeveloped, so they'll be in list form.
-Decision making requires at least some instinct, otherwise the animal wouldn't have an underlying goal.
-Humans think in terms of predicted senses, which are converted into associated neurotransmitters. That is part of the old brain, potentially.
-Besides Jeff Hawkins' idea of higher abstraction being achieved through every level creating higher-level abstractions, memories of ideas are important. We are more conscious about this type of abstraction. For example, knowing that 1+2=3 is known less consciously than knowing the quadratic formula. I think of this as sequentially-generated ideas, which are then remembered via electrical signals for a short amount of time. (You might not consciously know the quadratic formula starts with an x=.) This form of sequential consciousness would probably require a place to store things temporarily, like the hippocampus, although I could be wrong.
-The brain is a quantification machine. Reason, decision making, habits, and instinct are all forms of quantification.

More detailed posts:
#4: Early brain. It's not very affective/scientifically accurate, but it includes some old parts of the brain, and shows how their functions could've been used in the early brain.
#18: Sleep/how it is involved with learning. This is my favorite aspect of the brain.
#27: How the life cycle improves learning. This aspect of the brain is hierarchial and deals with quantifications, so I love it.
#34: The most fundamental way the brain learns.
#40: An alternative to naming neurons, learning cues to emotions, decision making
#41: Re-do of #40 (except naming neurons) and overall organization. I couldn't figure out how to incorporate prediction into the other method for decision making, so I found a new method. Click on the picture or use a magnifying glass.
#47: Re-do again of decision making, this time based directly on the brain. It relies on cravings, not happiness.

Originally Posted by NNet

These ideas are mainly for a neural network designed by Jeff Hawkins, called Hierarchial-Temporal Memory.
Here are some major topics which I think are important to the creation of brain-based machines. I'll explain more details when I have time, but since it will probably take me a while to write everything I want to, here is something more general.
Neurotransmitters affect large groups of neurons at once, in a variety of ways. Neurotransmitters can adjust thresholds, affect different types of neurons differently, or change the amount of time a neuron collects input before it can fire again. Neurotransmitters are highly involved in emotions, which are fundamentally an instinct to act a certain way (this idea is not entirely my own. Based on the Lovheim cube of emotions.). That could be seen as giving an artificial neural network a hint as to what it is meant to do. I think neurotransmitters create a hierarchy of goals, which helps an animal achieve the next highest quality of living, similar to Maslow's hierarchy of needs. Neurotransmitters are also important to learning, especially during sleep. While a person is asleep, new ideas can be explored without any danger. During sleep, the neocortex recieves random data from the pons, which could be like looking at a sequence of scribbles, and interpretting them as dreams. This is more effective because of the neurotransmitters involved in sleep. Besides adjusting plasticity and how quickly the brain changes focus, the neurotransmitters involved in sleep cause the neocortex to pay much less attention to detail, which is useful for interpretting random information without being distracted by noise.
The life cycle is important to development. As embryos, we learn the most basics laws of the universe, such as it is 3-d. As babies, we learn about food and comfort, and then speech. As children, we learn complex social skills. As adults, we learn to pay taxes and have a job. Imagine if a baby were given a job. It would be clueless, and it wouldn't learn anything. Instead of giving babies jobs, we slowly increase the complexity of their lives until they reach adulthood. That idea could be applied to neural networks.
The rest of these ideas are underdeveloped, so they'll be in list form.
-Decision making requires at least some instinct, otherwise the animal wouldn't have an underlying goal.
-Humans think in terms of predicted senses, which are converted into associated neurotransmitters. That is part of the old brain.
-Besides Jeff Hawkins' idea of higher abstraction being achieved through every level creating higher-level abstractions, memories of ideas are important. We are more conscious about this type of abstraction. For example, knowing that 1+2=3 is known less consciously than knowing the quadratic formula. I think of this as sequentially-generated ideas, which are then remembered via electrical signals for a short amount of time. (You might not consciously know the quadratic formula starts with an x=.) This form of sequential consciousness would probably require a place to store things temporarily, like the hippocampus, although I could be wrong.
-The brain is a quantification machine. Reason, decision making, habits, and instinct are all forms of quantification.

Have you ever heard about memristors.If not should you should study them.

Originally Posted by NNet

Have you ever heard about memristors.If not should you should study them.

Thanks, I will.

I guess I'll post all of my ideas on this thread, as I create them.

Recently I did some research and thinking on the basal ganglia, which is involved in motor functions and habits.
Here's some general data I found:
The basal ganglia has an instinct section and a habit-forming section. These sections are each composed of many nuclei, but I'll explain why that doesn't make a difference in another section. I'm not trying to re-design the basal ganglia, I'm trying to use them to create a better intelligence.

I have some basic evidence that this was the first form of learning, which I present below.

Here are some ideas:
1. The habit forming section is what actually outputs instincts to the rest of the brain, after a bit more processing. That processing includes deciding whether to output that instinct, as well as modifying the output.
2. Instincts are activated by cues. In animals on the level of nemotodes, the cues are smells or general shapes. (The general means a 3-d shape is a cue, but since vision is 2-d, some ORing circuitry is needed.) In more advanced animals, those cues can also be signals from other parts of the brain.
3. If those cues are of the simple type, the section of the basal ganglia which deals with habits/simple processing means that the animal will have a simple type of intelligence. It's somewhat hard to explain, but the idea is that instinct+habit forming-->learning.
4. There is no actual habits processing. Instead, the non-instinct part has neurons whose thresholds decrease whenever the neuron fires, and increases whenever the neuron doesn't fire for a little while. The processing is still there because of the random part of the learning algorithm explained below.

Consider a brain which has the following instincts:
-go towards x, y, and z smells.
-to swim, activate x and y muscles in x repeating sequence.
-adjust that sequence of muscle movement a bit every now and then. When x and y (good) smells are recieved, reduce plasticity, and when a and b (bad) smells are recieved, increase plasticity.
This method of adjusting plasticity is easy to implement, using neurotransmitters. One thing which must be done is to adjust the affect of the Hebbian algorithm or decrease thresholds. The later will cause it to use instincts for less extreme cues. It would also cause neurons, at a given moment, which would normally be active to activate, causing the neuron to form new connections. It would be like saying "Neurons which are almost active would be active if they had more connections, so we'll activate them so they'll form more connections".
The next thing that must be done is to make the input more random, which leads to experimenting. A simple instinct-type circuit could do this.
Once those changes are used during a period of experimentation, here's what will happen:
-Random changes will be made, some good and some bad. However, the animal won't be in danger if it fails to avoid predators, because it will be speeding along.
-Whenever the animal is near a good scent, the plasticity will drop. This means that the animal will have good sets of connections for longer periods of time than bad sets of connections. Because those connections will be reinforced more often than the bad connections, those connections will stick around.
That like seem like a very slow method of learning, but it isn't thanks to some awesome properties of Hebbian learning. Whenever something is done, it will become a habit, so it will be done more often, so it will become even more of a habit. At some point, neurons die if they fire too often, so this loop won't get out of control.

In summary, instinct+habits-->learning using this algorithm because good connections will be less plastic, so they will be dwelt on and form habits. It is a partially a genetic algorithm, so it has some flaws.
Evidence!:
Obviously, humans don't learn much in the way explained above. However, I think we evolved from something which uses this learning method. Here are some reasons why:
1. Humans have similar times of learning-sleep. Humans don't do much while they're asleep, whereas this idea means the animal will run around like a lunatic. However, both involve rapid learning and protection from harm while learning unreliable information (which is then made reliable while learning at normal speeds.)
2. Sleep in both humans and this involves changes in neurotransmitter levels. One function of sleep-related neurotransmitters in humans is to increase plasticity, just like in sleep-related neurotransmitters do this primative thingamajig.
3. The human brain has an ancient part called the pons. During sleep, it sends random data to the rest of the brain. In this primative thingamajig, it would distort the input the non-instinct section. (One thing I forgot to mention is that inputs would go to both the instinct and non-instinct parts, which means through the learning algorithm non-instinct behavior could arise.)
4. I don't know everything about the brain, so there are sure to be details which suggest that this idea is correct, as well as details which suggest that this idea needs tweaking.

So what can we use this for?, and how could we make this practical?
Jeff Hawkins' machine makes predictions, whereas my machine learns the best method of doing things.
Jeff Hawkins' idea is currently used to predict energy consumption, server loads, and he plans to use it to predict the weather.
My idea could be used to create little insect-like things. Because it uses instinct, ROM could be used for the instinct, and only a hundred or so neurons, or less depending on the use, would be used for the non-instinct part. Conventional artificial intelligence techniques could also be used for instinct. Or we could take a biological route and evolve a nemotode's brain into something else (nemotodes only have instinct, as far as I know.)
-We could mass produce microscopic robots, which float around in a substance and kill pathogens.
-We could use them to removed burnt cells, or cancer cells.
-We could attach enzymes to them, and use them to make chemical reactions happen very quickly.
If they weren't microscopic:
-Eat mosquitoes
-Eat garbage
The main purpose of these would be to get other machinery where it needs to go, like other simple animals, and then instruct that machinery on what to do.

Part of my central dogma is decision making, which requires instinct for the animal to have basic goals. Decision making requires both reasoning and instinct, so what if we were to combine this idea and Jeff Hawkins' idea?
The non-instinct part wouldn't be vital, but it would help the artificial neural net develop while young.. Some changes would have to be made and some additional circuits would have to be added, but decision making is a real possibity. With decision making, we can create intelligent robots (which aren't actually to useful), make non-bias government decisions, or do anything humans do without emotion.

What aspects of the central dogma does this relate to?
Primarily, neurotransmitters, sleep, decision making, and quantification.
Slightly development.
Edit: If I were to code something, I wouldn't code this. The reason I'm explaining this is because it has many similarities to the current brain, at least with my knowledge. One of my goals is to "evolve" a brain, or at least understand the fundamental parts. If this is really how primative life forms used to work, then there's not much to be learned from this. Luckily, I learned at least one useful thing from this. Intelligent parts of the brain are extremely different from simpler parts, because the simple parts are meant to cause the creature to survive, whereas the intelligent parts are more indirect.
Also, I'll probably use the idea of Hebbian habituation in the same program as my other ideas. I'll probably just use it so it can recognize unknown things in a different way than the neocortex, allowing for curiousity, to control attention.

Originally Posted by NNet

I guess I'll post all of my ideas on this thread, as I create them.

Recently I did some research and thinking on the basal ganglia, which is involved in motor functions and habits.
Here's some general data I found:
The basal ganglia has an instinct section and a habit-forming section. These sections are each composed of many nuclei, but I'll explain why that doesn't make a difference in another section. I'm not trying to re-design the basal ganglia, I'm trying to use them to create a better intelligence.

I have some basic evidence that this was the first form of learning, which I present below.

Here are some ideas:
1. The habit forming section is what actually outputs instincts to the rest of the brain, after a bit more processing. That processing includes deciding whether to output that instinct, as well as modifying the output.
2. Instincts are activated by cues. In animals on the level of nemotodes, the cues are smells or general shapes. (The general means a 3-d shape is a cue, but since vision is 2-d, some ORing circuitry is needed.) In more advanced animals, those cues can also be signals from other parts of the brain.
3. If those cues are of the simple type, the section of the basal ganglia which deals with habits/simple processing means that the animal will have a simple type of intelligence. It's somewhat hard to explain, but the idea is that instinct+habit forming-->learning.
4. There is no actual habits processing. Instead, the non-instinct part has neurons whose thresholds decrease whenever the neuron fires, and increases whenever the neuron doesn't fire for a little while. The processing is still there because of the random part of the learning algorithm explained below.

Consider a brain which has the following instincts:
-go towards x, y, and z smells.
-to swim, activate x and y muscles in x repeating sequence.
-adjust that sequence of muscle movement a bit every now and then. When x and y (good) smells are recieved, reduce plasticity, and when a and b (bad) smells are recieved, increase plasticity.
This method of adjusting plasticity is easy to implement, using neurotransmitters. One thing which must be done is to adjust the affect of the Hebbian algorithm or decrease thresholds. The later will cause it to swim faster, which is beneficial because it will avoid predatation more so. The animal will be moving faster, so it will also experience more smells during this time of experimentation. That's beneficial because it will be done experimenting sooner (the time between each new scent will be reduced, so in the same amount of time more data can be used to learn and thus a reliable amount of data can be collected in a short enough time period that there isn't a great risk of a major mistake in adjusting connections), and so be able to experiment more often.
The next thing that must be done is to make the input more random, which leads to experimenting. A simple instinct-type circuit could do this.
Once those changes are used during a period of experimentation, here's what will happen:
-Random changes will be made, some good and some bad. However, the animal won't be in danger if it fails to avoid predators, because it will be speeding along.
-Whenever the animal is near a good scent, the plasticity will drop. This means that the animal will have good sets of connections for longer periods of time than bad sets of connections. Because those connections will be reinforced more often than the bad connections, those connections will stick around.
That like seem like a very slow method of learning, but it isn't thanks to some awesome properties of Hebbian learning. Whenever something is done, it will become a habit, so it will be done more often, so it will become even more of a habit. At some point, neurons die if they fire too often, so this loop won't get out of control.

In summary, instinct+habits-->learning using this algorithm because good connections will be less plastic, so they will be dwelt on and form habits. It is a partially genetic algorithm, so it has some flaws, but it has methods to keep the creature safe while learning and I think it is elegant.

Evidence!:
Obviously, humans don't learn much in the way explained above. However, I think we evolved from something which uses this learning method. Here are some reasons why:
1. Humans have similar times of learning-sleep. Humans don't do much while they're asleep, whereas this idea means the animal will run around like a lunatic. However, both involve rapid learning and protection from harm while learning unreliable information (which is then made reliable while learning at normal speeds.)
2. Sleep in both humans and this involves changes in neurotransmitter levels. One function of sleep-related neurotransmitters in humans is to increase plasticity, just like in sleep-related neurotransmitters do this primative thingamajig.
3. The human brain has an ancient part called the pons. During sleep, it sends random data to the rest of the brain. In this primative thingamajig, it would distort the input the non-instinct section. (One thing I forgot to mention is that inputs would go to both the instinct and non-instinct parts, which means through the learning algorithm non-instinct behavior could arise.)
4. I don't know everything about the brain, so there are sure to be details which suggest this idea is correct, as well as details which suggest this idea needs tweaking.

So what can we use this for?, and how could we make this practical?
Jeff Hawkins' machine makes predictions, whereas my machine learns the best method of doing things.
Jeff Hawkins' idea is currently used to predict energy consumption, server loads, and he plans to use it to predict the weather.
My idea could be used to create little insect-like things. Because it uses instinct, ROM could be used for the instinct, and only a hundred or so neurons, or less depending on the use, would be used for the non-instinct part. Conventional artificial intelligence techniques could also be used for instinct. Or we could take a biological route and evolve a nemotode's brain into something else (nemotodes only have instinct, as far as I know.)
-We could mass produce microscopic robots, which float around in a substance and kill pathogens.
-We could use them to removed burnt cells, or cancer cells.
-We could attach enzymes to them, and use them to make chemical reactions happen very quickly.
If they weren't microscopic:
-Eat mosquitoes
-Eat garbage
The main purpose of these would be to get other machinery where it needs to go, like other simple animals, and then instruct that machinery on what to do.

Part of my central dogma is decision making, which requires instinct for the animal to have basic goals. Decision making requires both reasoning and instinct, so what if we were to combine this idea and Jeff Hawkins' idea?
The non-instinct part wouldn't be vital, but it would help the artificial neural net develop while young.. Some changes would have to be made and some additional circuits would have to be added, but decision making is a real possibity. With decision making, we can create intelligent robots (which aren't actually to useful), make non-bias government decisions, or do anything humans do without emotion.

What aspects of the central dogma does this relate to?
Primarily, neurotransmitters, sleep, decision making, and quantification.
Slightly development.

Why dont you try to make this into a a videogame to test out your idea.Make it into a virtual pet.But first you have to be a computer programmer.You could use the blender game engine and python.I have it it's open source.

Originally Posted by lightspeed

You could use the blender game engine and python.I have it it's open source.

My son is an aspie and he is using blender and gimp to create a 3d video game. he is doing pretty well at it but he does not know programming at all. I asked him just today what programming language his game was written in and he couldnt tell me. He just says blender does it all. I had to look up what language blender uses and found people in forums saying that it was c++ and/or python. I am trying to encourage my son to learn the actual programming so he can fine tune the aspects of his game but I can't get him to grasp the importance of knowing the basics. I am learning programming and know the basics of c++, a tiny bit of java, and web programming (html5, css, javascript).

Anyway, my son's game is pretty impressive considering he has no clue about programming at all.

Nnet, I looked up Numenta/jeff hawkins and it seems like it is a software/hardware product that is being promoted. Am I missing something?

I didn't see any links to the human brain in regards to hawkins/numenta.

I guess I should clarify what I am asking... is your idea something new? Because it seemed like common sense to me. (keep in mind i am not a neurobiologist or a highly skilled programmer, or even a psychologist for that matter)

Originally Posted by seagypsy

Originally Posted by lightspeed

You could use the blender game engine and python.I have it it's open source.

My son is an aspie and he is using blender and gimp to create a 3d video game. he is doing pretty well at it but he does not know programming at all. I asked him just today what programming language his game was written in and he couldnt tell me. He just says blender does it all. I had to look up what language blender uses and found people in forums saying that it was c++ and/or python. I am trying to encourage my son to learn the actual programming so he can fine tune the aspects of his game but I can't get him to grasp the importance of knowing the basics. I am learning programming and know the basics of c++, a tiny bit of java, and web programming (html5, css, javascript).

Anyway, my son's game is pretty impressive considering he has no clue about programming at all.

Nnet, I looked up Numenta/jeff hawkins and it seems like it is a software/hardware product that is being promoted. Am I missing something?

I didn't see any links to the human brain in regards to hawkins/numenta.

I guess I should clarify what I am asking... is your idea something new? Because it seemed like common sense to me. (keep in mind i am not a neurobiologist or a highly skilled programmer, or even a psychologist for that matter)

I am trying to get a hang of making ragdolls in the blender game engine.

Originally Posted by lightspeed

Originally Posted by seagypsy

Originally Posted by lightspeed

You could use the blender game engine and python.I have it it's open source.

My son is an aspie and he is using blender and gimp to create a 3d video game. he is doing pretty well at it but he does not know programming at all. I asked him just today what programming language his game was written in and he couldnt tell me. He just says blender does it all. I had to look up what language blender uses and found people in forums saying that it was c++ and/or python. I am trying to encourage my son to learn the actual programming so he can fine tune the aspects of his game but I can't get him to grasp the importance of knowing the basics. I am learning programming and know the basics of c++, a tiny bit of java, and web programming (html5, css, javascript).

Anyway, my son's game is pretty impressive considering he has no clue about programming at all.

Nnet, I looked up Numenta/jeff hawkins and it seems like it is a software/hardware product that is being promoted. Am I missing something?

I didn't see any links to the human brain in regards to hawkins/numenta.

I guess I should clarify what I am asking... is your idea something new? Because it seemed like common sense to me. (keep in mind i am not a neurobiologist or a highly skilled programmer, or even a psychologist for that matter)

I am trying to get a hang of making ragdolls in the blender game engine.

This is a sample of what my son is making. He just turned 18 a couple of days ago, but started using blender when he was 12.

[/QUOTE]Why dont you try to make this into a a videogame to test out your idea.Make it into a virtual pet.But first you have to be a computer programmer.You could use the blender game engine and python.I have it it's open source.[/QUOTE]
Thanks, that's an awesome idea which I plan to use eventually!
But not for a while. I started learning python this year, and I knew a bit about machine code before that, but I'm learning slowly. My main focus is designing stuff, so I'll probably start by writing a program for idle.
Oh, is blender easy to use? Awesome, I'll try that.

Originally Posted by seagypsy

Nnet, I looked up Numenta/jeff hawkins and it seems like it is a software/hardware product that is being promoted. Am I missing something?

I didn't see any links to the human brain in regards to hawkins/numenta.

I guess I should clarify what I am asking... is your idea something new? Because it seemed like common sense to me. (keep in mind i am not a neurobiologist or a highly skilled programmer, or even a psychologist for that matter)

Jeff Hawkins' idea is something new. It's based on a part of the neocortex which makes predictions (knowing if someone knocks at the door they will probably be there when you open the door), so it could predict power consumption server loads, etc. Because of the way that part of the neocortex works, it could also be used for vision or other sensory processing.
Numenta is a business, although it does research. It's a business so Numenta will have a larger effect on industry than it would otherwise.
Modeling Data Streams Using Sparse Distributed Representations - YouTube to learn how it works. It might be confusing, so feel free to ask any questions.
I think the biggest achievement this provides is that it is the first part of the brain humans truly understand, which opens many doors.
My own idea isn't nearly as original, but from the little I know, it works the same way as the brains of worms or other primative animals, so we could create a worm brain. Jeff Hawkins' idea is for the intelligence part of the brain, which is the most recent to evolve, whereas my idea is for one of the most primative parts of the brain.

Originally Posted by NNet

Originally Posted by seagypsy

Nnet, I looked up Numenta/jeff hawkins and it seems like it is a software/hardware product that is being promoted. Am I missing something?

I didn't see any links to the human brain in regards to hawkins/numenta.

I guess I should clarify what I am asking... is your idea something new? Because it seemed like common sense to me. (keep in mind i am not a neurobiologist or a highly skilled programmer, or even a psychologist for that matter)

Jeff Hawkins' idea is something new. It's based on a part of the neocortex which makes predictions (knowing if someone knocks at the door they will probably be there when you open the door), so it could predict power consumption server loads, etc. Because of the way that part of the neocortex works, it could also be used for vision or other sensory processing.
Numenta is a business, although it does research. It's a business so Numenta will have a larger effect on industry than it would otherwise.
Modeling Data Streams Using Sparse Distributed Representations - YouTube to learn how it works. It might be confusing, so feel free to ask any questions.
I think the biggest achievement this provides is that it is the first part of the brain humans truly understand, which opens many doors.
My own idea isn't nearly as original, but from the little I know, it works the same way as the brains of worms or other primative animals, so we could create a worm brain. Jeff Hawkins' idea is for the intelligence part of the brain, which is the most recent to evolve, whereas my idea is for one of the most primative parts of the brain.

Well you have definitely got my curiosity all in a tizzy. I am not knowledgeable enough on the subject to critique your idea or even know how original it is. It seems very similar to my own pondering, and since I don't consider myself to be all that smart I have a tendency to assume anything that matches my own imaginings is probably not that clever at all. So don't consider my opinion all that valid. It is based more on my low opinion of my own intellectual abilities than on yours. If it turns out you are a freaking genius then I may have to consider myself to be a tiny bit more clever than I assumed.


btw: It seems you accidently muddled up the quote attributions in your post. I noticed after posting my reply. I am correcting them in my post.

Originally Posted by seagypsy

Originally Posted by NNet

Originally Posted by seagypsy

Nnet, I looked up Numenta/jeff hawkins and it seems like it is a software/hardware product that is being promoted. Am I missing something?

I didn't see any links to the human brain in regards to hawkins/numenta.

I guess I should clarify what I am asking... is your idea something new? Because it seemed like common sense to me. (keep in mind i am not a neurobiologist or a highly skilled programmer, or even a psychologist for that matter)

Jeff Hawkins' idea is something new. It's based on a part of the neocortex which makes predictions (knowing if someone knocks at the door they will probably be there when you open the door), so it could predict power consumption server loads, etc. Because of the way that part of the neocortex works, it could also be used for vision or other sensory processing.
Numenta is a business, although it does research. It's a business so Numenta will have a larger effect on industry than it would otherwise.
Modeling Data Streams Using Sparse Distributed Representations - YouTube to learn how it works. It might be confusing, so feel free to ask any questions.
I think the biggest achievement this provides is that it is the first part of the brain humans truly understand, which opens many doors.
My own idea isn't nearly as original, but from the little I know, it works the same way as the brains of worms or other primative animals, so we could create a worm brain. Jeff Hawkins' idea is for the intelligence part of the brain, which is the most recent to evolve, whereas my idea is for one of the most primative parts of the brain.

Well you have definitely got my curiosity all in a tizzy. I am not knowledgeable enough on the subject to critique your idea or even know how original it is. It seems very similar to my own pondering, and since I don't consider myself to be all that smart I have a tendency to assume anything that matches my own imaginings is probably not that clever at all. So don't consider my opinion all that valid. It is based more on my low opinion of my own intellectual abilities than on yours. If it turns out you are a freaking genius then I may have to consider myself to be a tiny bit more clever than I assumed.


btw: It seems you accidently muddled up the quote attributions in your post. I noticed after posting my reply. I am correcting them in my post.

If you have thought of enough things to have happened to think of something I thought of, you must be smart. (Edit: if you happened to think of what I did, you must have thought of a lot of things, so...)
I'm not a freaking genious at all. I'm new to this field, although I use a different (biological-based) method than most.
Do you study neural networks or psychology, or ponder either?
Thanks for fixing the quotes, I have no idea how that works.

Originally Posted by NNet

Originally Posted by seagypsy

Originally Posted by NNet

Originally Posted by seagypsy

Nnet, I looked up Numenta/jeff hawkins and it seems like it is a software/hardware product that is being promoted. Am I missing something?

I didn't see any links to the human brain in regards to hawkins/numenta.

I guess I should clarify what I am asking... is your idea something new? Because it seemed like common sense to me. (keep in mind i am not a neurobiologist or a highly skilled programmer, or even a psychologist for that matter)

Jeff Hawkins' idea is something new. It's based on a part of the neocortex which makes predictions (knowing if someone knocks at the door they will probably be there when you open the door), so it could predict power consumption server loads, etc. Because of the way that part of the neocortex works, it could also be used for vision or other sensory processing.
Numenta is a business, although it does research. It's a business so Numenta will have a larger effect on industry than it would otherwise.
Modeling Data Streams Using Sparse Distributed Representations - YouTube to learn how it works. It might be confusing, so feel free to ask any questions.
I think the biggest achievement this provides is that it is the first part of the brain humans truly understand, which opens many doors.
My own idea isn't nearly as original, but from the little I know, it works the same way as the brains of worms or other primative animals, so we could create a worm brain. Jeff Hawkins' idea is for the intelligence part of the brain, which is the most recent to evolve, whereas my idea is for one of the most primative parts of the brain.

Well you have definitely got my curiosity all in a tizzy. I am not knowledgeable enough on the subject to critique your idea or even know how original it is. It seems very similar to my own pondering, and since I don't consider myself to be all that smart I have a tendency to assume anything that matches my own imaginings is probably not that clever at all. So don't consider my opinion all that valid. It is based more on my low opinion of my own intellectual abilities than on yours. If it turns out you are a freaking genius then I may have to consider myself to be a tiny bit more clever than I assumed.


btw: It seems you accidently muddled up the quote attributions in your post. I noticed after posting my reply. I am correcting them in my post.

If you have thought of enough things to have happened to think of something I thought of, you must be smart.
I'm not a freaking genious at all. I'm new to this field, although I use a different (biological-based) method than most.
Do you study neural networks or psychology, or ponder either?
Thanks for fixing the quotes, I have no idea how that works.

My fascination lies more in psychology but you can't avoid neurobiology when dealing with psychology unless you want to delude yourself that the mind is somehow separate from the brain. But I don't have any formal training in neurobiology and only low level education in psychology. The bulk of what I "know" about psychology is from my own observation, analysis and casual reading of psychology books and articles. I have too much experience dealing with mentally ill people to avoid trying to understand what I am dealing with.

I am also into programming, though not highly skilled yet. I am a beginner but the idea of how computers process information amazes me and knowing that we are nothing but organic machines susceptible to the same laws of physics, chemistry, and such as any other element in the universe tells me that our brains can't operate much differently than computer processors do. We are made of atoms which make up elements which exist throughout the universe whether as part of a living organism or not. We are made of metal, gases, and minerals. We are a glob of non living elements put together in a pattern that creates a mass that displays attributes of life. Nothing more, nothing less.

Why dont you try to make this into a a videogame to test out your idea.Make it into a virtual pet.But first you have to be a computer programmer.You could use the blender game engine and python.I have it it's open source

Thanks, that's an awesome idea which I plan to use eventually!
But not for a while. I started learning python this year, and I knew a bit about machine code before that, but I'm learning slowly. My main focus is designing stuff, so I'll probably start by writing a program for idle.
Oh, is blender easy to use? Awesome, I'll try that.

It is and the python programming language is a easy to learn.

That is petry good texturing seagypsy did he use texture paint to do that or just uv unwraped the models?

Originally Posted by seagypsy

Originally Posted by NNet

Originally Posted by seagypsy

Originally Posted by NNet

Originally Posted by seagypsy

Nnet, I looked up Numenta/jeff hawkins and it seems like it is a software/hardware product that is being promoted. Am I missing something?

I didn't see any links to the human brain in regards to hawkins/numenta.

I guess I should clarify what I am asking... is your idea something new? Because it seemed like common sense to me. (keep in mind i am not a neurobiologist or a highly skilled programmer, or even a psychologist for that matter)

Jeff Hawkins' idea is something new. It's based on a part of the neocortex which makes predictions (knowing if someone knocks at the door they will probably be there when you open the door), so it could predict power consumption server loads, etc. Because of the way that part of the neocortex works, it could also be used for vision or other sensory processing.
Numenta is a business, although it does research. It's a business so Numenta will have a larger effect on industry than it would otherwise.
Modeling Data Streams Using Sparse Distributed Representations - YouTube to learn how it works. It might be confusing, so feel free to ask any questions.
I think the biggest achievement this provides is that it is the first part of the brain humans truly understand, which opens many doors.
My own idea isn't nearly as original, but from the little I know, it works the same way as the brains of worms or other primative animals, so we could create a worm brain. Jeff Hawkins' idea is for the intelligence part of the brain, which is the most recent to evolve, whereas my idea is for one of the most primative parts of the brain.

Well you have definitely got my curiosity all in a tizzy. I am not knowledgeable enough on the subject to critique your idea or even know how original it is. It seems very similar to my own pondering, and since I don't consider myself to be all that smart I have a tendency to assume anything that matches my own imaginings is probably not that clever at all. So don't consider my opinion all that valid. It is based more on my low opinion of my own intellectual abilities than on yours. If it turns out you are a freaking genius then I may have to consider myself to be a tiny bit more clever than I assumed.


btw: It seems you accidently muddled up the quote attributions in your post. I noticed after posting my reply. I am correcting them in my post.

If you have thought of enough things to have happened to think of something I thought of, you must be smart.
I'm not a freaking genious at all. I'm new to this field, although I use a different (biological-based) method than most.
Do you study neural networks or psychology, or ponder either?
Thanks for fixing the quotes, I have no idea how that works.

My fascination lies more in psychology but you can't avoid neurobiology when dealing with psychology unless you want to delude yourself that the mind is somehow separate from the brain. But I don't have any formal training in neurobiology and only low level education in psychology. The bulk of what I "know" about psychology is from my own observation, analysis and casual reading of psychology books and articles. I have too much experience dealing with mentally ill people to avoid trying to understand what I am dealing with.

I am also into programming, though not highly skilled yet. I am a beginner but the idea of how computers process information amazes me and knowing that we are nothing but organic machines susceptible to the same laws of physics, chemistry, and such as any other element in the universe tells me that our brains can't operate much differently than computer processors do. We are made of atoms which make up elements which exist throughout the universe whether as part of a living organism or not. We are made of metal, gases, and minerals. We are a glob of non living elements put together in a pattern that creates a mass that displays attributes of life. Nothing more, nothing less.

The main focus of what I posted was the algorithm for learning.

Originally Posted by lightspeed

That is petry good texturing seagypsy did he use texture paint to do that or just uv unwraped the models?

All I know is he uses gimp to paint his objects. He has been living with his dad for the past 4 years 1000 miles away from me. And so I haven't been there to see what he how he is producing things.

When he was twelve I contacted WETA workshop to find out what kind of education he would need to work for them and they advised me that no degree was needed and that he could start learning cgi immediately. They suggested that he download blender and gimp since they are free and to have him play with those since Maya (the software they use) is very similar to blender. And gimp is a good replacement for most photoeditors that are expensive.

Then a few years later, when I started taking IT courses I had a teacher tell me to get him Alice so he can learn logic. So I did that and shortly after he got the hang of alice he started producing his game. But at that was just before he went to live with his dad. He watches tutorials for stuff on youtube but other than that he has no formal training of any kind on any type of software. He learns mostly by trial and error. If he had formal training he would be scary good.

Sleep
You’ll need to know how neurons work to understand a few tiny pieces of this. You might want to use wikipedia instead. Let's Build a Neural Net ep. 1-How NNs Work - YouTube
(As always, the data I'll mention here isn’t accurate. Don’t write anything I say here on an exam.)
Sleep has many functions, such as restoring the immune system and healing. The function I’ll focus on here is brain activity during sleep, especially neurotransmitter levels. Brain activity during sleep is geared towards learning.
Sleep is divided into stages. During each stage, brain activity levels are in waves of differing frequency. This is a very rough generalization, and during each stage brain activity constantly changes:
1: 11-16 hertz (brain waves per seconds)
2: 0.5-2 hertz at least 1/5 of the time, becoming more frequent over time, and high amplitude (slower and more synchronized, synchronized meaning more neurons at once.)
3: 11-16 hertz
4: REM sleep (over 13 hertz, with more memorable dreams than the other stages)
5: Cycle repeats
So brain waves become slower, then faster, then slower. I think the reason for this is that during sleep the brain experiments with new ideas—decisions, connections between memories, etc.—and tests these new ideas.
But how does it generate these new ideas to test? I don’t know for sure, but it might have something to do with one of the earliest parts of the brain- the pons. During sleep, the pons sends random data to the cortex. Random date may sound pointless, but perhaps it is like looking at scribbles and trying to see faces. The cortex interprets each new scribble it sees based on the previous scribble (because the cortex warps data towards what it thought would come next.)
To make this process more effective, neurotransmitters are used. Although there are many neurotransmitters, I’ll focus on a couple key neurotransmitters.
Serotonin:
Serotonin levels are high during all stages, except for REM sleep. It is excitatory to neurons in the cortex, which is highly beneficial. Before I explain why high serotonin levels are beneficial, it is necessary to explain how the cortex learns.
The cortex learns using a Hebbian algorithm, which means that active neurons connect to some other active neurons (makes those neurons more likely to activate when the neuron is on.) This creates knowledge. Each neuron represents an attribute of what the brain is sensing, so connecting neurons which are directly activated by inputs to higher level neurons (represent higher abstractions) adds knowledge. This is somewhat hard to explain, so if you don’t understand this rambling either trust me or look up Jeff Hawkins/Numenta on Youtube.
Now that that’s been explained, let’s return to serotonin, which is excitatory to neurons in the cortex. Unlike some neurotransmitters, serotonin is distributed to large areas of the brain (whereas some neurotransmitters are exchanged between single neurons.) I like to think of this as changing the properties of many neurons at once. I think serotonin simply lowers the threshold of every neuron during sleep. This means that neurons which would be almost active without the serotonin will be active with the serotonin, and so connect to other neurons. This is like saying “Neurons which are almost active would probably be active if they knew more connections, so we’ll activate nearly-active neurons.” This is a form of exploratory learning, which could be a detriment rather than a benefit if used during wakefulness.
Acetylcholine:
Acetylcholine levels are high during REM sleep, when the exploratory changes of non-REM sleep are fine-tuned. Like serotonin, acetylcholine is a “wide-area neurotransmitter”. However, it affects neurons in a more complex way than serotonin. Rather than affecting all neurons in one way, it affects different types of neurons differently.
The neurons which act as inputs to the neocortex are inhibited (their thresholds are increased) by acetylcholine. This is like filtering out weak inputs. Acetylcholine also increases the depolarization time of neurons, meaning that acetylcholine makes neurons collect data for a longer time before firing. The overall effect of high acetylcholine levels is the ignorance of details.
By ignoring the many details of random input from the pons, more realistic dreams are produced. This fine tunes neurons by testing the exploratory changes of non-REM sleep in more realistic dreams.
Reasons for different brain wave frequencies during different stages of sleep:
Brains waves during non-REM sleep are low frequency and highly synchronized. This means that neurons will be activated by more neurons at a time, when a brain wave occurs. This has a similar effect to serotonin. There are probably other purposes. For example, more synchronization means the random dreams of non-REM sleep will cause less problems. (That’s sort of hard to explain without explaining how the cortex works.)
Brain waves during REM sleep are high frequency and less synchronized. Synchronization isn’t necessary, because neurons shouldn’t be activated more heavily than normal. Because synchronization isn’t necessary, parts of the cortex can go at their own speed, increasing the frequency of brain waves. In addition, dreams will be more realistic, so synchronization isn’t as important.
What can we use these ideas for?:
These ideas could be applied to Grok, a software program designed by Numenta which makes predictions. A period of inactivity would be necessary, which isn’t a good thing for something meant to constantly receive data. If Grok were made fast enough, data could be collected while Grok sleeps, then sent to Grok later on. Another option is to use two Groks, so one can make predictions while the other sleeps, and send its predictions to the other Grok once it wakes. Overall, these ideas could make Grok learn faster and learn more abstract ideas.

You have a subconscious mind and a conscious mind how does that factor into your idea for ai?Because some people don't remember their dreams.

Originally Posted by lightspeed

You have a subconscious mind and a conscious mind how does that factor into your idea for ai?Because some people don't remember their dreams.

I don't really know the difference, nor do I think there really is one for the part of the brain I'm talking about. During sleep, changes are mainly for factual knowledge (concepts like 1+1=72). Although people experience dreams, those dreams are difficult to remember probably because they are usually expected. Dreams are produced based on what the person knows facts-wise, so it is rare for a dream to develop to the point of complexity that the person didn't expect them. The hippocampus only remembers unexpected things/is where most episodic (memories of events) are stored.

In order for ai to be any good it really has to learn.It's not what put into the computer that counts it's what comes out.This how i would make ai i would make it have a imagination.I would make it capable of imitation as well.It would be capable of recognising the difference between similar and disimilar patterns.In it's environment.Just a few of my thoughts but not all of them.It must experience pleasure and unpleasant things.It must feel in order to learn.

Originally Posted by lightspeed

In order for ai to be any good it really has to learn.It's not what put into the computer that counts it's what comes out.This how i would make ai i would make it have a imagination.I would make it capable of imatation as well.It would be capable of recognising the difference between similar and disimilar patterns.In it's environment.Just a few of my thoughts but not all of them.

Thanks for sharing your ideas, I would appreciate more if you want to share them.
This is a neural network, meaning it works the same way as the brain. To recognize patterns, it finds what things usually occur together, which is very simple to do with neurons and effective. Imagination is via the ability for it to use its outputs as inputs.
Jeff Hawkins' hierarchial-temporal memory, based on the neocortex, is far more useful than anything else produced by A.I. It is actually inelligent, rather than giving specific outputs for certain inputs.

But first you need to define what would be pleasant and what would be unpleasant in order for it to really learn.It could be randomly selected from information it has learned.

Originally Posted by lightspeed

But first you need to define what would be pleasant and what would be unpleasant in order for it to really learn.It could be randomly selected from information it has learned.

The really awesome thing about Hawkins' hierarchical temporal-memory is that it doesn't need to have that. It makes predictions about what will happen next, so on its own it is useful. In the human brain, things are defined as good/bad senses, but the part I focus on doesn't require that. It just finds connections between things.

Unpleasant would be anything that invalidates a previously held concept of reality that is determined to reflect a pattern that predicts proper functioning of the perceiver. Pleasant would be anything that validates the existing that theory of reality.

However if the current concept of reality is one that predicts harm to the one perceiving it and new data invalidates that prediction then this new data would be pleasant.

So.. um.. ok .. pleasant would be any data that contributes to a prediction of proper function of the one perceiving it. unpleasant would be data that causes a prediction of harm (improper function)to the one perceiving it.

Originally Posted by seagypsy

Unpleasant would be anything that invalidates a previously held concept of reality that is determined to reflect a pattern that predicts proper functioning of the perceiver. Pleasant would be anything that validates the existing that theory of reality.

However if the current concept of reality is one that predicts harm to the one perceiving it and new data invalidates that prediction then this new data would be pleasant.

So.. um.. ok .. pleasant would be any data that contributes to a prediction of proper function of the one perceiving it. unpleasant would be data that causes a prediction of harm (improper function)to the one perceiving it.

Oh. It does excactly that, and some other original stuff (Jeff Hawkins' HTM)
Edit: Kind of. It notices when it's predictions are wrong, and adjusts them, basically.

Life cycle:
Learning is done differently during each stage of the life cycle. Before I explain each stage, I should mention a few things. If the brain gets confused, it won't learn much. Less developed brains get confused more easily, so they should be given less advanced tasks. In education, kids are given more and more complex problems as they age, building on the problems which they could already solve. This idea could be applied to artificial neural nets.
1: Embryo. (Learning type: 3-D universe.) Many think that infants are born with a cortex which knows nothing. This is certainly not true. The embryo stage of the life cycle is when the brain learns the simplest patterns of the universe. For example, they learn that the universe is 3-d, or how to control movement. (Aside: means embryos could be transfered to a higher dimensional universe, and they would function normally.)
In addition, this stage is a highly experimental stage. The brain starts with many more neurons than it needs, and then it removes most neurons later on. I believe that neurons which don't fire often are removed, because they haven't been able to learn much due to disfunctional internals. This could be applied to manufacturing of artificial organic neural nets, by allowing a chance of disfunction for each neuron. Another reason for starting with extra neurons could be that the neurons start with random connections, like random guesses which might be close to correct connections, or might be far from correct connections. If the neuron fires often, it is close to a correct pattern.
That was kind of long for one stage, so here's a summary: the embryo stage starts with extra neurons. Some die due to poor manufacturing. Some die because the connections they started with where highly incorrect (this process actually continues throughout life), and this means the manufacturing of artificial organic neural nets doesn't have to be very effective. In the end, only neurons with decent starting connections are left. During this stage, embryos also learn their most basic knowledge, which underlies the rest of their learning.
2: Baby. (Learning type: omnomnomnomnomnomnom food, eeeewwww chalk.) Babies don't have a developed cortex, nor do they face the complex issues of later life, such as paying taxes. Plasticity is increased during this stage, and babies experiment a lot.
3: Children. (Learning level: 1+43-56x45-8=2) Now that the cortex knows basic information such as what to do to achieve something, more complex learning can occur. Children are introduced to many subject during this stage, both academic and social.
Plasticity is lower now, because this stage determines the rest of the kid's life. Information learned during the embryo stage and baby stage will be consistent from brain to brain, but for the first time now, a variety of things could be learned. If plasticity were higher, the child would learn too quickly and miss important patterns.
4: Teen. (Learning type: essays, more complex math.) During this stage, plasticity is increased. The brain now has a lot of bredth of knowledge, so it is safe to learn quickly.
During this stage, I think emotional intelligence develops a lot. Teens have strong emotions, so they learn how to operate with intense emotions. They also learn about life goals, which are related to emotions.
5: Everything else, a continuum of learning in college, then much self-actualization.
One thing I'd like to note is that stages 1 and 2 are akin to stages 3 and 4. They're a period of low plasticity and preliminary learning, then a period of high plasticity and in-depth learning. This could be applied to artifical neural networks, potentially.
This is a hierarchy which deals with quantification . The hierarchy is an increase of complexity of knowledge. This is involved with quantification because it allows the brain to make quantifications at every stage of deveopment, by preventing confusion.
Like all hierarchies of the brain, each level has individual differences, but the hierarchy has a general pattern.
Wow, I was planning for this to be 1/8 its length. I learn a lot while writing these. By writing everything in one place, it lets me make more quantifications.

Could the  neural network of jeff hawkins do what a child prodigy does naturely?

Could the neural network of jeff hawkins do what a child prodigy does naturely?

Originally Posted by lightspeed

Could the neural network of jeff hawkins do what a child prodigy does naturely?

Nope, but they can do other cool stuff. For example, they can have any type of sense, whether or not humans have them. Built of the same materials as computers, they could be thousands of times faster than us.

Originally Posted by NNet

Originally Posted by lightspeed

Could the neural network of jeff hawkins do what a child prodigy does naturely?

Nope, but they can do other cool stuff. For example, they can have any type of sense, whether or not humans have them. Built of the same materials as computers, they could be thousands of times faster than us.

Why don't they test it out in a robot like HRP-4C?It's a type of android model.I would like to see androids capable of having sex.They have made real dolls why can't they make the real thing?

Originally Posted by lightspeed

Why don't they test it out in a robot like HRP-4C?It's a type of android model.I would like to see androids capable of having sex.They have made real dolls why can't they make the real thing?

Not sure I'd mess around with an android that has a designation that reads like "HeRPes-4C..."

I recently found out that the pons doesn't send random data to the cortex during NREM, and that the data might not be completely random during REM (I'll have to look into that, but if it is semi-random, the idea still works ish.)
Here's the new idea for NREM:
-serotonin plays the same role
-NREM is a period of thinking without any distractions, rather than dreaming.

Here’s a simplified description of Hebbian learning and why it works.
Designing neural networks is like making soup. You add various things to the mix, which mix together to make good soup. Obviously, there is some kind of base. For example, tomato soup could be the base. After we’ve got the base, we add whatever else we want—onions, potatoes, salt, or whatever goes in soup. (I’ve never made soup.)

The base:
Hebbian learning is surprisingly simple. Whenever a neuron fires, it connects to other neurons which are firing at the same time. To paraphrase a Wikipedia article, “neurons which fire together wire together.” Connecting means the firing of one neuron makes the other more likely to fire at the same time.

Ingredient #1—Attribute Representation
Neurons represent an attribute of something. For example, if you were looking at a dog, a neuron which represents “living” would fire, as well as a neuron which represents “brown”. Together, all the firing neurons when you see a dog represent “dog”. I should note that neurons typically represent something much simpler. A neuron can represent anything from an input from the retina, to a corner, and all the way up to perception and thought.

Ingredient #2—Hierarchy
The neurons are arranged in layers, stacked like pancakes. The lowest layer receives the input, and then each layer sends data to the next one up.
Each layer contains neurons which represent more complex attributes than the last. This is because neurons can only connect to neurons in the next highest layer.

The point is difficult to explain, so I’ll use an example. Let’s say neuron A is one layer lower than neuron B. Neuron A represents a head, and neuron B represents a dog. Whenever we see a dog, neuron B will fire because it represents a dog. Neuron A will also fire, because dogs have legs. Because they will usually both fire at the same time, neuron A will connect to neuron B, and we’ll have learned that dogs have legs.

The last issue is naming neurons, or deciding what each neuron means. This process is difficult to explain, and it’s not necessarily true. I looked up everything else online, so everything else is probably used by the brain.

The rest of the soup is a lot easier to make. We could add neurotransmitters, sleep, development, or anything else we want.

If anything is confusing, please tell me.

Nnet you did not tell me about opening a thread for your beloved jeff hawkins. Anyways am not going to read through the posts.Cheers.

Originally Posted by NNet

Here’s a simplified description of Hebbian learning and why it works.
Designing neural networks is like making soup. You add various things to the mix, which mix together to make good soup. Obviously, there is some kind of base. For example, tomato soup could be the base. After we’ve got the base, we add whatever else we want—onions, potatoes, salt, or whatever goes in soup. (I’ve never made soup.)

The base:
Hebbian learning is surprisingly simple. Whenever a neuron fires, it connects to other neurons which are firing at the same time. To paraphrase a Wikipedia article, “neurons which fire together wire together.” Connecting means the firing of one neuron makes the other more likely to fire at the same time.

Ingredient #1—Attribute Representation
Neurons represent an attribute of something. For example, if you were looking at a dog, a neuron which represents “living” would fire, as well as a neuron which represents “brown”. Together, all the firing neurons when you see a dog represent “dog”. I should note that neurons typically represent something much simpler. A neuron can represent anything from an input from the retina, to a corner, and all the way up to perception and thought.

Ingredient #2—Hierarchy
The neurons are arranged in layers, stacked like pancakes. The lowest layer receives the input, and then each layer sends data to the next one up.
Each layer contains neurons which represent more complex attributes than the last. This is because neurons can only connect to neurons in the next highest layer.

The point is difficult to explain, so I’ll use an example. Let’s say neuron A is one layer lower than neuron B. Neuron A represents a head, and neuron B represents a dog. Whenever we see a dog, neuron B will fire because it represents a dog. Neuron A will also fire, because dogs have legs. Because they will usually both fire at the same time, neuron A will connect to neuron B, and we’ll have learned that dogs have legs.

The last issue is naming neurons, or deciding what each neuron means. This process is difficult to explain, and it’s not necessarily true. I looked up everything else online, so everything else is used by the brain.
The rest of the soup is a lot easier to make. We could add neurotransmitters, sleep, development, or anything else we want.

If anything is confusing, please tell me.

I think neurons decide their names by what energys they are exposed to and the energy that is received by the brain is filtered through the brain many layers of neurons.Your brain develops in response to it's environment.

Originally Posted by merumario

Nnet you did not tell me about opening a thread for your beloved jeff hawkins. Anyways am not going to read through the posts.Cheers.

I don't think I opened this until after we were done arguing, and I still haven't really explained how Hawkins' HTM works.
I think Jeff Hawkins is a cocky jerk. I was joking about him being beloved. I do not think of him as a king, but as someone with awesome ideas which I'd like to build on.
I doubt more than a few people have really read what I wrote. I personally would not have read my posts, because they're long and confusing (which is something I'd like to change.)
I respect your beliefs, but I don't think a phsycology section is the appropriate place to talk debate them. I'm open to religious ideas, and I am very slightly religious. I just don't think the soul is seperate from the brain, but rather a result of it's structure.
As seagypsy said, I shouldn't have kept arguing with your beliefs. I'm sorry.

@lightspeed
That's a good point, neurons are named simply based on the data they recieve. The problem is causing neurons to have names which represent something, and activating the neurons whenever their aspect is true. The biggest problem is determining what each neuron actually means at the output, and determining what the implications of those are.

to be deleted

Here’s a method to name neurons, generate emotions based on what is perceived, and decide what to do.

When the brain first develops, each neuron starts with some random connections. Those random connections define what each neuron means. The problem is that each neuron will activate at times inconsistent with what it actually means, because it could mean “meow” yet be associated with “dog”. Whenever you hear a meow, you would be more likely to think of a dog, not a cat. However, this problem is soon remedied. Every time you hear a meow, you see a cat. The Hebbian aspect of learning means that you will begin to associate “meow” with “cat,” and not with “dog”. So neurons start as a slightly incorrect name (“dog” = “legs” and “big black wet nose” and “meow”), and are adjusted to be a correct name. Any connections inconsistent with the overall meaning of the other connections are destroyed, and new connections are formed to create a more precise meaning.

The next issue is determining what to do based on the meaning of the output. For example, if the output means “jungle cat,” you should avoid it. If the output means “4+3,” you should quantify that into “7”.

In this method, the actual names of neurons are not determined. Instead, their implications are determined. For example, “4” and “+” and “3” has the implication “7,” which is learned because those attributes are associated with 7. (Recall Hebbian learning- associating things with other things, based on when they are an attribute of what is being perceived.)

This type of implication is entirely within the part of the brain which I talk about most- the cortex, which is the intelligence part of the brain. This type of implication is pretty much what I explained in the last post. The point is that the neurons of the cortex don’t know what the names of other neurons mean. They simply connect based on Hebbian learning.

There is another type of implication- determining “good” or “bad,” “happy” or “sad”. The limbic system generates emotions (I think). In order to determine which emotion to generate, the limbic system must know the implications of what is perceived. For example, “drinking soda” has the implication “happy”. I think that (in this case) this implication is not learned. It is an implication you are born knowing, because neurons already encode that knowledge. Those neurons have fixed connections, so soda will always trigger the release of serotonin, unless other parts of the brain have learned that soda is icky because it makes you feel sick.

This type of implication can also be learned. For this example, we’ll use “reading a book”. When we are born, we do not like reading books. However, we do like learning. We’re curious. Once we learn how to read, we start to like reading books. Reading any given sentence of a paragraph wouldn’t lead to learning, but reading the entire paragraph would. Reading at any given moment indirectly leads to learning, so we learn to like reading even at non-learning moments.

To learn “reading a book” should make you happy, both the limbic system and the cortex are involved. Together, they learn that “reading a book” is often true at the same time as “learning”. To learn this, the neurons (of the highest layer) of the cortex that represent “reading a book” and the neurons of the limbic system that represent “learning” are connected. Both groups will be on at the same time, so simple Hebbian learning is enough to learn the connection between “reading a book” and “learning”. Whenever you read a book, the neurons which represent “learning” will activate and make you happy.

The last thing I’d like to talk about is decision making.
Everyone is born knowing to do what makes them happy. Doing what makes you happy is just another implication. However, this implication is different from the non-learned implications I talked about before. Instead of being from the cortex to the limbic system, it is an implication from the limbic system to the cortex.

When you’re making a decision, your cortex imagines possible actions. For example, you see some soup. Through some amazing properties of the cortex (which I will explain in a later post), your cortex imagines “eating the soup with a fork”, “eating the soup with a knife,” or “eating the soup with a spoon”. As you imagine each story, your limbic system becomes happy or sad. Once you reach a story which makes you happy, your limbic system releases dopamine. Dopamine is the neurotransmitter of motivation, or deciding to do something. When the cortex receives the dopamine, it stops imagining and starts doing. It snaps back to reality and, with the idea of “use a spoon to drink the soup” in mind, it causes you to use a spoon to drink the soup.

A couple questions to ask myself/notes:
-I’m not sure how the thing currently being imagined would be done. If you’ve already imagined the whole story, would keeping it in mind cause you to do what you decided?
-How does dopamine cause the cortex to stop imagining?
-The limbic system might not be the emotional center. It is involved in memory (which theoretically is similar to the cortex), as well as other things. Alternatively, both the limbic system and the brain stem might be involved in emotion, but I think that the part of the brain stem I’m talking about is just for neurotransmitters.

That got a little rambly near the end. If you have any questions, please ask. I like answering questions.

Decision Making
Decision making requires basic goals. Everything you do is based on a couple basic goals—minimizing negative stimuli, and maximizing positive stimuli. Those two goals are really a bunch of little goals, such as drinking, not getting burnt, being successful, etc. These basic goals are custom created for each neural network, so a neural net could play checkers, control a maze to confuse a player, monitor a computer for failure, or whatever else.
The main issue is causing the cortex to try to achieve those goals, using prediction. The brain uses a reward and reinforcement system, but I don’t understand how it works in terms of wiring, so I created my own method. It definitely isn’t the best possible, but it works. Here’s the organization:
bandicam 2013-01-21 10-43-45-966.jpg

This looks like an incomprehensible diagram. And it is, without further explaining.
Cortex/Mental State: (Pre-Made)
The cortex/mental state is what I explained in the previous sections. It is the largest part, and learns to be intelligent. The mental state is the highest layer of the cortex, so it is technically part of the cortex.

Limbic System: (User-Made)
The neural network’s limbic system isn’t really a system, unlike the real limbic system. It is a group of a few dozen neurons which determine how happy the net is. Each neuron represents a reason for being happy. If a particular net played checkers, a neuron would represent capturing a piece, another would represent losing a piece, and another would represent getting a king. Each would cause a shift in the level of happiness.

Instinct: (User-Made)
The Instinct circuit contains two pieces. The first sends data to the limbic system, by activating specific neurons. For each input set, it activates certain neurons which represent reasons for being happy or sad. It defines what those neurons mean (except for the shift in happiness, which is calculated within the limbic system.)
The instinct module can also create outputs. For example, if it would be able to win in tic-tac-toe, it would create the output which would cause the net to win.

Input/Output (User-Made)
The input will usually be simple wires in. Some pre-processing could be added. For example, a similar method to both the thalamus and perceptrons could be used. There would be two layers. The first layer would be the input layer, and the second layer would be composed of neurons. Each input would activate a few other random neurons, which would make the cortex learn more quickly.
The output is similar. It will usually be simple wires out, but some post-processing could be added. In the brain, the cerebellum makes sure there are no conflicting muscle commands. A similar circuit could be added to the output, which could make sure the output isn’t horribly wrong.

Putting the Pieces Together:
So far, you probably don’t understand how this works, unless you read ahead.
I like to think of the mental state as an advanced view of the world. Rather than seeing a grid of pixels like the instinct circuit, the mental state sees objects and causes.
Both the mental state and instinct module have inputs to both the limbic system and output. However, the instinct module uses fixed connections, and the mental state uses Hebbian connections. The instinct basically trains the mental state’s connections, because the mental state’s signals must correspond to the instinct’s signals, generally. This is true for the mental state’s output to the limbic system and the output.
That means the cortex will do things which make sense for the given goals. The main problem is using prediction to make a better decision. The cortex sends predictions to the limbic system, which essentially allows the limbic system to predict how happy the net will be, including because of the current mental state. This means that the net’s happiness at a given moment takes into account how happy it will be in the future, so all that must be done
The mental state sends data to the output, so the current mental state determines the action. There are aspects (one per neuron) of the current mental state. Some make the neural net happy, and some make the neural net sad.
Each “happy neuron” of the limbic system connects to the neurons in the mental state which are active when it is active. In addition, each “sad neuron” does the same, but sends inhibitory signals. The overall effect is exciting good choices for the output, and inhibiting bad choices for the output.
Each aspect of the output represents something like a muscle command, a specific move (in which case the output would be user-designed to select the most heavily active output.) It is like the highest level of the cortex (it is like a mental), so it is used in the system of prediction as the cortex.

NOTE: The limbic system sends data to the output, rather than the mental state. Also, most of this text is copied from another website which is for a game rather than science, so there might be some strange naming I forgot to change.

bandicam 2013-02-10 16-03-08-497.jpg
This idea is based on the idea that dopamine increases levels of imagination. I'll edit this post to be more detailed later. Click the picture for a larger size.

Originally Posted by NNet

bandicam 2013-02-10 16-03-08-497.jpg
This idea is based on the idea that dopamine increases levels of imagination. I'll edit this post to be more detailed later. Click the picture for a larger size.

Has there been an experiment proving that dopamine increases the levels of imagination.If it does do this then their might be a drug that decreases level of the imagination.

Originally Posted by NNet

This idea is based on the idea that dopamine increases levels of imagination.

Hmm,

Originally Posted by NNet

One possibility is that dopamine in the cortex causes less imagination.

Dopamine

Originally Posted by Dywyddyr

Originally Posted by NNet

This idea is based on the idea that dopamine increases levels of imagination.

Hmm,

Originally Posted by NNet

One possibility is that dopamine in the cortex causes less imagination.

Dopamine

Yeah, kind of obssessed. I posted here because I have a functional system now.

Originally Posted by lightspeed

Originally Posted by NNet

bandicam 2013-02-10 16-03-08-497.jpg
This idea is based on the idea that dopamine increases levels of imagination. I'll edit this post to be more detailed later. Click the picture for a larger size.

Has there been an experiment proving that dopamine increases the levels of imagination.If it does do this then their might be a drug that decreases level of the imagination.

Not exactly. It is definetly inhibitory, decreasing (not increasing) cortical activity. The question is whether or not it decreases predictive signals. (Note that lower cortical dopamine levels lead to increased impulsivity, so I think dopamine levels are decreased in response to a potential reward. Increasing predictive signals would have this affect, because the cortex actually predicts what in will do in order to do something, generally.)

I think you got it right. I'm not sure what schizophrenia medicine does. It changes dopamine levels, presumably to be more normal/higher in the cortex.
One thing I love about neuroscience is that so many ideas are connected. If one aspect of the brain suggests one way the brain functions, the other aspects will usually suggest that as well.

Here are the basic principles of this system. Thinking of dopamine as excitatory here is easier:
1. Dopamine is released in response to a potential reward. ^
2. If you see a chocolate bar as a baby, you might eat it because of instinct only. This causes "seeing chocolate bar" to cause you to predict "eating chocolate bar". Seeing a chocolate bar also increases your dopamine levels, causing you to predict "eating a chocolate bar" more. This causes the cortex to make you eat a chocolate bar, because prediction is the same as thinking something is true presently, only less so (although you can learn to make the distinction between prediction and what is currently happening.) This idea is from Jeff Hawkins- the cortex controls what you do with prediction alone, sending data down its hierarchy to unfold high-level ideas into little details of muscle movements.
3. This system is based on "cravings" rather than happiness.
4. It is important for the cortex to have an effect on the neurons which produce dopamine, so you can become motivated to go to work, which isn't part of your instinct.

The rest of this was intended to be sent to Hawkins, so it might be a bit confusing. I didn't feel like writing a better version. If things conflict, use the newer ideas (lines above this one.)

Goals are required for behavior. Everything a person does is an attempt to achieve a set of basic goals encoded within their brain.
Simply encoding the goals isn’t enough. The cortex must try to achieve those goals.
There are several mechanisms which cause the brain to try to achieve its goals. The reward system is major, but I don’t know how rewards work. Rather than creating rewards and reinforcing behavior, you could cause something similar to cravings, and reinforce behavior using an indirect method which I will explain. This idea relies heavily on prediction.
Here are some vital mechanisms:
1. Whenever a goal is achieved, dopamine is produced. Dopamine is also produced when a goal is imagined to be achieved (which is more important in this article). This is easy to do, because neurons in their predictive state can be considered active, and treated the same as neurons in their normal active state. This is extremely important for dopamine’s role in this system.
2. Dopamine increases the strength of imagination signals, and decreases the strength of normal signals. If dopamine levels are high enough, your brain will keep imagining an action being considered, and probably execute the action.
3. Prediction causes the brain to try to achieve secondary goals (which will aid the actual, primary, goals in the future), because imagining executing a secondary goal causes the brain to imagine executing a primary goal, which essentially makes it motivated or unmotivated.
4. Norepinephrine has the opposite effect of dopamine. It causes you to think about something else, by affecting the thalamus perhaps. It is also excitatory.

I’ll analyze some parts and most connections of the picture here, to explain how this system works in more detail.
Norepinephrine + Dopamine Neurons
Each neuron of this part represents a reason to be motivated (dopaminergic neurons) or a reason to be unmotivated (norepinephrinergic neurons). It isn’t necessary for neurons in this part to interact.
Goals Encoded
This part encodes the goals. It activates dopaminergic or norepinephrinergic neurons based on the input alone. It doesn’t take into account data from the cortex, which is done through a different mechanism.
Norepinephrine + Dopamine Neurons to Cortex
This connection controls how powerful the imagination connections are. It doesn’t send signals to specific neurons. Instead, it modulates the activity of all cortical neurons with dopamine and norepinephrine. The level of each neurotransmitter depends on how many neurons of that neurotransmitter are active.
Cortex to Norepinephrine + Dopamine Neurons
This connection is vital for learning secondary goals. Recall that the brain doesn’t know what each output of the cortex means, but instead uses an indirect method.
This method is difficult to explain, but it is simple once understood. The N. and D. neurons will activate at the same time as certain neurons in the highest cortical level (which send outputs), because activity of both parts generally corresponds to certain inputs. This fact means that the highest-level neurons will send connections to the N. and D. neurons which should be excited by what the neuron represents, because of Hebbian learning.
Norepinephrine also reduces the thresholds of cortex neurons. I’ll explain why while I explain the next connection.
This method causes the brain to try to achieve primary goals, and therefore secondary goals. You can think of the “Goals Encoded” as representing the goodness of the current input, and the cortex output as being the future input, which is essentially analyzed by the “Goals Encoded” as well, but indirectly.
Norepinephrine + Dopamine Neurons to Output
This connection makes the brain less able to do something when dopamine levels are low. If there is enough cortical activity, then the brain will still output a signal.
Norepinephrine doesn’t make a difference here. When norepinephrine levels are high, cortical activity increases, so the brain can think of a way to solve the problem. If both dopamine and norepinephrine levels are high, the brain should be able to execute the action. Dopamine doesn’t represent happiness; it represents a good idea. Norepinephrine represents both negativity of an idea and the need to find a good but experimental idea.
Cortex to Output
It probably seems strange to use Hebbian connecting between the cortex and output. This is effective because of other parts of the brain. When the brain is motivated, the output neurons are more likely to be active, and neurons of the mental state connect to active output neurons. This means cortical neurons will cause a beneficial output.
Note that starting with random connections would be useful, though unnecessary, because neurons connect to random neurons Hebbianly, not all of them.
To help speed up this process, output neurons could Hebbianly connect to each-other. A circuit similar to a region of cortex could even be used, so simultaneous movements would be quickly learned, and patterns could be quickly learned.

Overall, this method means the brain tries to achieve certain goals, both through primary goals and secondary goals which it creates itself.
This section was confusing. All of the mechanisms should work well, except for the output method. It isn’t very effective, so I’ll look for a different method.