Dude you got the degree I wanted and seemingly living the life of my dreams, good going! Awesome! Aerospace engineer that builds artificial hardware neurons on breadboard in his pasttime, that is so cool, subscribed because this is 1:1 my interests.
@@GlobalScienceNetwork I am by no means an expert, but definitely someone with passion for the fields involved in it, hardware SNN's interest me a bunch!
Very cool. To make it more similar a biologic neuron we need: -autoreceptors; ex; on a excitatory cell released neurotransmitter can bind to the post synaptic receptor which depolarize the post synaptic cell and cause an action potential, but it can also bind to pre-synaptic receptors to inhibit further release of transmitter. Autoreceptors/pre synaptic rec. are thus inhibitory. The cell/neuron will continously alter its expression of autoreceptors which in turn regulate how likely it is to propagate the signal to the next cell. Whether we actually need autoreceptors on artificial neurons idk, but as opposed to 2 neurons acting on a 3rd post synaptic neuron where the combined effect is either inhib. or stim., autoreceptors add the ability of the excitatory neuron becoming "more" or "less" excitatory depending on various input (hormones, etc) This is only needed in artificial neurons if we want f.ex to simulate "feelings"; i.e when "scared" the excitatory neuron has a more powerful stim. effect than when "calm".
Interesting. We defiantly want to simulate "feelings" as this is part of life. I bet we could add this type of change into the circuit with a "light depending resistor" or something similar that changes based on some needed feeling parameter. Any other thoughts on how to build this in?
Hi, This is very cool, you created a novel dynamic system that self activates! It's also aesthetically pleasing. As usual, I have a few questions and comments: 1. Is the diode really necessary since I don't think you need to protect against negative voltages? Maybe its purpose is to simulate an activation threshold at the "knee". 2. Is the self-activating pattern somewhat stable and deterministic, or somewhat random and chaotic? It's hard to tell from the video, even when slowing it down. Since transistors vary as they heat up, I would imagine there would be some variation. 3. Does the system form a complete loop (i.e. "recurrent")? In other words is the "first" neuron itself triggered by another neuron or is it just set to a fixed potential? 4. I could see that some configurations might settle out, stop cycling and reach a stable pattern. Have you experienced that? Have you determined the conditions for continous activation? Did you have to "tweak" it to get a stable active pattern? 5. BTW, I like your neuron video, did you produce that yourself? It's a very good animation! (Sorry I have so many questions 😊) Thanks!
Good questions as usual. 1. The diode is needed so that when the synapse is off charge does not flow back to the ground. 2. Well the next video will explain how the neuron itself works. This video was just for the synapse. Since you asked though, the neurons are made with a Schmitt trigger used as a comparator to determine when the voltage is above a threshold value. The charge is flowing into the capacitor based on the resistor values. When the voltage in the capacitor is above the threshold the capacitor discharges. So I would say it is deterministic. 3. The input current going into the first neurons will be from some sensor. Right now it is just set based on the resistor value used. The neurons downstream are connected and are affected by all synapses that are connected to it. Each synapse can drive the voltage within the capacitor to be closer or further from the required threshold voltage to discharge. Depending on if it is an excitatory or inhibitory synapse. 4. As long as there are more excitatory synapses than inhibitory synapses feeding into the neuron it should fire in a continuous manner. In some cases, the inputs might be such that the neuron will not fire. Overall though the entire network will be continuously firing just like in the brain. 5. Thanks! The animations were made in Maya. I tried making them but there was quite a learning curve for Maya so I hired someone to make the animations. The animations are custom made though just for these videos.
@@GlobalScienceNetwork Thanks for answering my questions. BTW, the diode-capacitor pattern is commonly known/used as a "peak detector". Again, nice production, pretty serious that you hired someone out for the video! I look forward to your upcoming videos!
I think that everyone who starts working with neural networks should start with this. I may be old-fashioned, but it’s hard for me to take seriously anyone who considers himself an "expert" in the field of neural networks, having never held a transistor in his hands and has no idea about the structure of a biological prototype.
I think that a lot of the neural network modelling is based on calculating values for variables in a mesh network based on their weights, much like the number of receptors activated to trip a synapse. But I do think a neural computer is a more interesting idea.
Yeah, it is difficult to build a circuits that are comparable to the human body/brain, that is why we have not done it yet. Neuroscientist are sill trying to map all the connections in the brain and they have a long way to go. It is the only science project worth working on in my opinion though as the results have real meaning.
I followed the setup at 5:25 but the blue light seems to turn on without me pressing the button. Could the resistors be the cause? If so, could you please tell me what the values are? I'm having a hard time seeing the colors (and still learning how to read schematics). Thanks
Hi At 6:00 I show the transistor level schematic. The resistor values are shown but your issue is likely that you did not add a diode before the 10K potentiometer which is a variable resistor. You do not need to use the potentiameter and can just use a 2K resistor but you do need to have the diode in place.
Polarizing material between the optocouplers? You can do a lot of bioanalogous things with the raw optocoupler from a pair of LED's thing, you could even do volume transmission. Also opamps and memristors work great for creating spiking neurons.
What do you mean by "Polarizing material between the optocouplers"? Do know of a good example/demo where opamps and memristors are used to create spiking neurons?
@@GlobalScienceNetwork I found some examples on researchgate that checked out in falstad about two years ago, I'll have to go diving. I thought of something akin to polarized lenses, you can take two of those and have one rotate on a servo (very crude example) and use that to make an adjustable filter, LED's of different colours are sensitive to different forms of light, mostly from the plastic "bulb" being dyed. I had something that can be polarized electronically in mind but I forgot the actual name of the thing.
how do you think this can model the brain's plasticity? I mean, how can these artificial synapses rearrange themselves "at runtime" based on learning patterns?
Well the textbook answer would likely be to use memristors. Basically these reduce the resistance with more activity. In reality though this is not a great model for how the brain actually learns. It is up to us to develop better hardware based synapses and neurons. The specificity of charge distribution is actually really important. I will talk about this more in upcoming videos. Basically if there are 5,000 incoming synapses only 50 of those connections may be from the output of another neuron yet it will fire in sequence. This is because of the location of the connections. Also when a synapse needs to be inhibitory for the logic to prevent the neuron from firing this can be done by changing just a few synapses at the right locations. In reality it may be that most neurons want to be firing at a higher rate but it is the inhibitory synapses preventing that. So by changing those few synapses it really lets you control that neurons firing rate. That is my thought at the moment.
I wish you could simplify the circuit diagram because I'm facinated by this project and wish to explore this area of science. Can you include an article where all information can be sourced including the circuit diagram in discrete components not in modules
I am glad that you are interested in the project! If I did not show the circuit diagram with discrete components in this video, I should have in the artificial neuron video. When the diagram is shown at the logic gate level, you can see how to build each type of logic gate in the how to build logic gate video. It is good to start learning the logic gates first. Let me know if you need any help when you are trying to build the circuits!
Great. I try to provide all the information you need to build them within the videos. If you have any specific questions, let me know, and I will try to help.
Yeah an LED can work like a solar cell, although it is not very efficient as the band gap of the material is higher than silicon's, 1.1 eV , which is typically used in solar cells. The small current that is generated by the LED is then amplified by using two transistors as an amplifier. I just used them cause I had the LEDs and the circuit would be similar regardless of the receiving component. You can also use photodiodes, phototransistors, or LDRs for the receiving side of the trigger.
I do not think the LED would be damaged, I have not had any stop working. The main thing that breaks LEDs is to much current. There is a current limiting resistor in the circuit. In a final design the LED would be Infared LED emitting IR light on to a photo diode or photo transistor. The LED work good though for a demonstration how the system works.
To build the artificial synapse and neuron you need 10 2N2222 NPN transistors, a white LED, Yellow LED, blue LED, diode, variable resistor, 3 100K resistors, 220 ohm resistor, 3 2K resistors, 1 10K resistors, a 47uF capacitor, breadboard, 5V battery pack, and wire. If you watch the artificial neuron video I show the circuit diagram at the component level.
That would work for a basic neuron. To build neurons that learn and change over time we will need memristor, resistors, capacitors, and carefully designed synapses . It will be involved but it is good to start with a simple design and see how far we can go with that. It should be fun!
The light would add some charge to the receiving side. If we use the optocouplers that are in the IC packaging the outside light and light from other synapses would be blocked. It is a good point though that if we used a sensitive receiver in an open environment like this it could add charge to other synapses. That could actually be a design feature rather than an issue if we wanted one output to contact many other inputs without having to wire them all together.
@@GlobalScienceNetwork Or allow a limited number of other neuron inputs to be coupled to a single (or more) neuron on a computational level. The problem as I see it is being able to get adjustable light signal levels from each input independently with minimal circuitry. :)
Ha ha this made me laugh! There are several paradox's in information theory that propose an individual can not duplicate enough information to create itself. So I am trying to get others to be interested in these types of projects!
@@GlobalScienceNetwork I am interested just gonna have to keep trying I’m more of a written word person if that makes any sense I haven’t developed a good enough aptitude..
Having tinkered with transistors 50 years ago, it is nice to see how you are able to bridge the gap from that time to the dawn of artificial intelligence.
Dude you got the degree I wanted and seemingly living the life of my dreams, good going! Awesome! Aerospace engineer that builds artificial hardware neurons on breadboard in his pasttime, that is so cool, subscribed because this is 1:1 my interests.
Sweet! Yeah defiantly follow along and I bet you will be able to help out! I should have some more good content rolling out soon.
@@GlobalScienceNetwork I am by no means an expert, but definitely someone with passion for the fields involved in it, hardware SNN's interest me a bunch!
Very cool.
To make it more similar a biologic neuron we need:
-autoreceptors; ex; on a excitatory cell released neurotransmitter can bind to the post synaptic receptor which depolarize the post synaptic cell and cause an action potential, but it can also bind to pre-synaptic receptors to inhibit further release of transmitter. Autoreceptors/pre synaptic rec. are thus inhibitory. The cell/neuron will continously alter its expression of autoreceptors which in turn regulate how likely it is to propagate the signal to the next cell.
Whether we actually need autoreceptors on artificial neurons idk, but as opposed to 2 neurons acting on a 3rd post synaptic neuron where the combined effect is either inhib. or stim., autoreceptors add the ability of the excitatory neuron becoming "more" or "less" excitatory depending on various input (hormones, etc)
This is only needed in artificial neurons if we want f.ex to simulate "feelings"; i.e when "scared" the excitatory neuron has a more powerful stim. effect than when "calm".
Interesting. We defiantly want to simulate "feelings" as this is part of life. I bet we could add this type of change into the circuit with a "light depending resistor" or something similar that changes based on some needed feeling parameter. Any other thoughts on how to build this in?
I am starting to think you are not joking in you intro
Ha ha yeah, we are starting with the basics but I was not joking! Building neural networks with hardware is the future!
In the end, hardware is better and more advanced than software and is more reliable. Although I do not prefer the hardware from electronics type.
This is everything i want to study. Thank you
Hi,
This is very cool, you created a novel dynamic system that self activates! It's also aesthetically pleasing. As usual, I have a few questions and comments:
1. Is the diode really necessary since I don't think you need to protect against negative voltages? Maybe its purpose is to simulate an activation threshold at the "knee".
2. Is the self-activating pattern somewhat stable and deterministic, or somewhat random and chaotic? It's hard to tell from the video, even when slowing it down. Since transistors vary as they heat up, I would imagine there would be some variation.
3. Does the system form a complete loop (i.e. "recurrent")? In other words is the "first" neuron itself triggered by another neuron or is it just set to a fixed potential?
4. I could see that some configurations might settle out, stop cycling and reach a stable pattern. Have you experienced that? Have you determined the conditions for continous activation? Did you have to "tweak" it to get a stable active pattern?
5. BTW, I like your neuron video, did you produce that yourself? It's a very good animation!
(Sorry I have so many questions 😊)
Thanks!
Good questions as usual.
1. The diode is needed so that when the synapse is off charge does not flow back to the ground.
2. Well the next video will explain how the neuron itself works. This video was just for the synapse. Since you asked though, the neurons are made with a Schmitt trigger used as a comparator to determine when the voltage is above a threshold value. The charge is flowing into the capacitor based on the resistor values. When the voltage in the capacitor is above the threshold the capacitor discharges. So I would say it is deterministic.
3. The input current going into the first neurons will be from some sensor. Right now it is just set based on the resistor value used. The neurons downstream are connected and are affected by all synapses that are connected to it. Each synapse can drive the voltage within the capacitor to be closer or further from the required threshold voltage to discharge. Depending on if it is an excitatory or inhibitory synapse.
4. As long as there are more excitatory synapses than inhibitory synapses feeding into the neuron it should fire in a continuous manner. In some cases, the inputs might be such that the neuron will not fire. Overall though the entire network will be continuously firing just like in the brain.
5. Thanks! The animations were made in Maya. I tried making them but there was quite a learning curve for Maya so I hired someone to make the animations. The animations are custom made though just for these videos.
@@GlobalScienceNetwork Thanks for answering my questions. BTW, the diode-capacitor pattern is commonly known/used as a "peak detector". Again, nice production, pretty serious that you hired someone out for the video! I look forward to your upcoming videos!
I think that everyone who starts working with neural networks should start with this. I may be old-fashioned, but it’s hard for me to take seriously anyone who considers himself an "expert" in the field of neural networks, having never held a transistor in his hands and has no idea about the structure of a biological prototype.
I agree 100 percent! If you follow along I bet you will be able to help improve these designs!
I think that a lot of the neural network modelling is based on calculating values for variables in a mesh network based on their weights, much like the number of receptors activated to trip a synapse. But I do think a neural computer is a more interesting idea.
Way above my pay grade. Kudos
Thanks!
our human body is way more advance then we think ,its really amazing to learn all these
Yeah, it is difficult to build a circuits that are comparable to the human body/brain, that is why we have not done it yet. Neuroscientist are sill trying to map all the connections in the brain and they have a long way to go. It is the only science project worth working on in my opinion though as the results have real meaning.
What if electronic circuits are not the solution ? What if the solution was more natural and not consuming ?
I followed the setup at 5:25 but the blue light seems to turn on without me pressing the button. Could the resistors be the cause? If so, could you please tell me what the values are? I'm having a hard time seeing the colors (and still learning how to read schematics). Thanks
Hi At 6:00 I show the transistor level schematic. The resistor values are shown but your issue is likely that you did not add a diode before the 10K potentiometer which is a variable resistor. You do not need to use the potentiameter and can just use a 2K resistor but you do need to have the diode in place.
Polarizing material between the optocouplers? You can do a lot of bioanalogous things with the raw optocoupler from a pair of LED's thing, you could even do volume transmission.
Also opamps and memristors work great for creating spiking neurons.
What do you mean by "Polarizing material between the optocouplers"? Do know of a good example/demo where opamps and memristors are used to create spiking neurons?
@@GlobalScienceNetwork I found some examples on researchgate that checked out in falstad about two years ago, I'll have to go diving. I thought of something akin to polarized lenses, you can take two of those and have one rotate on a servo (very crude example) and use that to make an adjustable filter, LED's of different colours are sensitive to different forms of light, mostly from the plastic "bulb" being dyed. I had something that can be polarized electronically in mind but I forgot the actual name of the thing.
how do you think this can model the brain's plasticity? I mean, how can these artificial synapses rearrange themselves "at runtime" based on learning patterns?
Well the textbook answer would likely be to use memristors. Basically these reduce the resistance with more activity. In reality though this is not a great model for how the brain actually learns. It is up to us to develop better hardware based synapses and neurons. The specificity of charge distribution is actually really important. I will talk about this more in upcoming videos. Basically if there are 5,000 incoming synapses only 50 of those connections may be from the output of another neuron yet it will fire in sequence. This is because of the location of the connections. Also when a synapse needs to be inhibitory for the logic to prevent the neuron from firing this can be done by changing just a few synapses at the right locations. In reality it may be that most neurons want to be firing at a higher rate but it is the inhibitory synapses preventing that. So by changing those few synapses it really lets you control that neurons firing rate. That is my thought at the moment.
I wish you could simplify the circuit diagram because I'm facinated by this project and wish to explore this area of science. Can you include an article where all information can be sourced including the circuit diagram in discrete components not in modules
I am glad that you are interested in the project! If I did not show the circuit diagram with discrete components in this video, I should have in the artificial neuron video. When the diagram is shown at the logic gate level, you can see how to build each type of logic gate in the how to build logic gate video. It is good to start learning the logic gates first. Let me know if you need any help when you are trying to build the circuits!
I wanna try building it ... Just to learn more about it ...
Can you gelp me with some reference or anything else?
Great. I try to provide all the information you need to build them within the videos. If you have any specific questions, let me know, and I will try to help.
How is the receiving LED producing enough signal to trigger anything. I didn't know LEDs could work that way.
Yeah an LED can work like a solar cell, although it is not very efficient as the band gap of the material is higher than silicon's, 1.1 eV , which is typically used in solar cells. The small current that is generated by the LED is then amplified by using two transistors as an amplifier. I just used them cause I had the LEDs and the circuit would be similar regardless of the receiving component. You can also use photodiodes, phototransistors, or LDRs for the receiving side of the trigger.
Ty.
Sure thing!
But, aren't the LED damaged by turning on and off repeatedly ?
I do not think the LED would be damaged, I have not had any stop working. The main thing that breaks LEDs is to much current. There is a current limiting resistor in the circuit. In a final design the LED would be Infared LED emitting IR light on to a photo diode or photo transistor. The LED work good though for a demonstration how the system works.
🎉Guao eres un genio🎉
Ja ja ¡Gracias!
component list?
To build the artificial synapse and neuron you need 10 2N2222 NPN transistors, a white LED, Yellow LED, blue LED, diode, variable resistor, 3 100K resistors, 220 ohm resistor, 3 2K resistors, 1 10K resistors, a 47uF capacitor, breadboard, 5V battery pack, and wire. If you watch the artificial neuron video I show the circuit diagram at the component level.
I think you could just use a capacitor with a bunch of variable resistors as a neuron.
That would work for a basic neuron. To build neurons that learn and change over time we will need memristor, resistors, capacitors, and carefully designed synapses . It will be involved but it is good to start with a simple design and see how far we can go with that. It should be fun!
The incident light in the room would act as an enhancing function to all neurons. Interesting.
The light would add some charge to the receiving side. If we use the optocouplers that are in the IC packaging the outside light and light from other synapses would be blocked. It is a good point though that if we used a sensitive receiver in an open environment like this it could add charge to other synapses. That could actually be a design feature rather than an issue if we wanted one output to contact many other inputs without having to wire them all together.
@@GlobalScienceNetwork Or allow a limited number of other neuron inputs to be coupled to a single (or more) neuron on a computational level. The problem as I see it is being able to get adjustable light signal levels from each input independently with minimal circuitry. :)
Bro gonna develop AGI all by himself..
Ha ha this made me laugh! There are several paradox's in information theory that propose an individual can not duplicate enough information to create itself. So I am trying to get others to be interested in these types of projects!
@@GlobalScienceNetwork I am interested just gonna have to keep trying I’m more of a written word person if that makes any sense I haven’t developed a good enough aptitude..
@@henrylawrence5566 That makes sense. If you are interested in these types of projects I bet you would be surprised what you could build!
I need to see that put on a piece of silicone to see what it does ❤❤❤❤❤❤❤
Yeah, eventually I plan to make that a reality!
Wow
Thank you!
Having tinkered with transistors 50 years ago, it is nice to see how you are able to bridge the gap from that time to the dawn of artificial intelligence.
from the above video we conclude that we can build ann artifical oragnism and theN A COMPLEX ORGANISM
Yeah, we can try and see if it works!
Dude I thought you just joking
Ha ha I was being serious! There will be some more good content coming soon!
imagine , this is the first thing our children do when you give them a kit
Ha ha it actually would be a fun first circuit board project!
Suggestion: try to vary your tone of voice. Thank you otherwise for the video.
Ok, thanks!
You literally milked the synapse clip. 😄
Ha ha well it is important!
dude you have built artificial synapes0 on breadboards just create something new make games on breadboards using transistors.
Games would be cool as well but I think the artificial brains are cool as well! It will eventually become the most widely used/manufactured tech.