I have in the past wrote a softwre that simulate a neural network based on Hebb model. Means that electrical impulsions processed by a neuron (physically speaking) could be translated bya mathematical function (works with threshold function) that takes impulsion as input and calculate another impulsion as output. The model I have designed translated quite well the basic behaviour of neurons in the brain. After this step I have decided to add many neurons to design a software model that represent a network. I faced many problems because many data were required. I am wandering if in neuroscience you guys have models of neurons that behave jointly and/or each neuron by itself. That could help in thinking of a first approach. Proposals for a common work is welcome. The purpose is obviousely to design a primitive model of a neural network that simulate training or memorization.
Perhaps I am more up to date on NN design than some of the rest of you.
The Hebbian model was used to design the Perceptron, in the 1970's Marvin Minksy noted the lack of second order functions in his book "Perceptron" which created the "Hidden Layer" model. Another approach was the HH model based on Hodgkin & Huxleys ground breaking "Spiking" neuron research. At this point, the scientists were still working with giant squid neurons, and the "Learning Function" was based on the delta model, where an "Error" value was propagated through the network to change the weights on the individual neurons. Work by Alkon on Snails created the concept of a pre-synaptic/post-synaptic adjustment mechanism, Which allowed the neuron to learn both by hebbian learning (Strengthening the link between neurons acting at the same time) and Feedback related to the firing of the neuron. Kandel's work on snails pointed out the link between synaptic values, and gene expression, suggesting a role for synapses in expression of the genes necessary for dendrite/axon growth.
Sessions, of McGill University, suggested that there might be multiple chemical pathways involved in memory, each having its own genetic expression, and its own memory term, that acted together to create the single output. In order to implement this, I proposed the idea that a protein "Flag" could be associated with a synapse to trigger different stages of memory within the same cell. I broke these flags up into:
1. Temporary sequesteration, triggering ion channel replacement, or adding new ion channels to an existing synapse, but where the protein was absorbed by the membrane reconstruction mechanism, and
2. Survivalble Flags where the flag was retained as part of the synapse, and triggered secondary chemistry that in turn released a secondary messenger each time the synapse was activated, which in turn activated certain chemical channels associated with later memory stages within the cell.
One concept of mine, is that the 40hz signal expressed by EEG triggers in Cortex Tissue, the S synapse which in turn triggers Long-term memory growth. this would be consistent with Kandels NMDA synapse in the hippocampus that is thought to have the same role, since they share calcium channels, and secondary messengers, but the cortex tissues would operate according to shorter timespans, and operate with facilitation in order to sustain signals for about 3 seconds, while the NMDA synapses would work with LTP with terms of up to 2 weeks.
I don know how could i help you, we studying HH (Hodgkin Huxley) and BVP model in electronics circuit. Also we designing small world network of neurons using electronic circuits.
To my opinion, you work falls in a new/old direction of bioinformatics. Decades ago, people started to use computer to mimic/decode intelligent activities macroscopically with EEG and microscopically with microelectrode array (MEA). Particularly for MEA, people need to deal with a large quantity of field potential firing. In that kind of research, a lot of software has been designed to deal with electrical activities at different network levels. You might want to check into that kind of literature. Since it has been a research area for a quite while, there are many books on this. The one I have is sort of old. It was got from Amazon. The title is "Toward Replacement Parts for the Brain: Implantable Biomimetic Electronics as Neural Prostheses". I think the current computer technology can let you integrate much more information and enable a great in-depth computational research than before. Good luck!
Perhaps I am more up to date on NN design than some of the rest of you.
The Hebbian model was used to design the Perceptron, in the 1970's Marvin Minksy noted the lack of second order functions in his book "Perceptron" which created the "Hidden Layer" model. Another approach was the HH model based on Hodgkin & Huxleys ground breaking "Spiking" neuron research. At this point, the scientists were still working with giant squid neurons, and the "Learning Function" was based on the delta model, where an "Error" value was propagated through the network to change the weights on the individual neurons. Work by Alkon on Snails created the concept of a pre-synaptic/post-synaptic adjustment mechanism, Which allowed the neuron to learn both by hebbian learning (Strengthening the link between neurons acting at the same time) and Feedback related to the firing of the neuron. Kandel's work on snails pointed out the link between synaptic values, and gene expression, suggesting a role for synapses in expression of the genes necessary for dendrite/axon growth.
Sessions, of McGill University, suggested that there might be multiple chemical pathways involved in memory, each having its own genetic expression, and its own memory term, that acted together to create the single output. In order to implement this, I proposed the idea that a protein "Flag" could be associated with a synapse to trigger different stages of memory within the same cell. I broke these flags up into:
1. Temporary sequesteration, triggering ion channel replacement, or adding new ion channels to an existing synapse, but where the protein was absorbed by the membrane reconstruction mechanism, and
2. Survivalble Flags where the flag was retained as part of the synapse, and triggered secondary chemistry that in turn released a secondary messenger each time the synapse was activated, which in turn activated certain chemical channels associated with later memory stages within the cell.
One concept of mine, is that the 40hz signal expressed by EEG triggers in Cortex Tissue, the S synapse which in turn triggers Long-term memory growth. this would be consistent with Kandels NMDA synapse in the hippocampus that is thought to have the same role, since they share calcium channels, and secondary messengers, but the cortex tissues would operate according to shorter timespans, and operate with facilitation in order to sustain signals for about 3 seconds, while the NMDA synapses would work with LTP with terms of up to 2 weeks.
- Badges
- Science topic
- Similar topics
- Biological Science
- Neuroscience