Is there research that employs realistic models of neurons? Usually, the model of a neuron for a neural network is quite simple as opposed to the realistic neuron, which involves hundreds of proteins and millions of molecules (or even greater numbers). Is there research that draws implications from this reality and tries to design realistic models of neurons?

Particularly, recently, Rosehip neuron was discovered. Such neuron can be found only in human brain cells (and in no other species). Are there some implications for neural network design and operation that can be drawn by realistically modelling this Rosehip neuron?


3 Answers 3


State of Rosehip Research

The Rosehip neuron is an important discovery, with vast implications to AI and its relationship to the dominant intelligence on earth for at least the last 50,000 years. The paper that has spawned other articles is Transcriptomic and morphophysiological evidence for a specialized human cortical GABAergic cell type, Buldog et. al., September 2018, Nature Neuroscience.

The relationship between this neuron type and its DNA expression is beginning. No data is available regarding the impact of the Rosehop distinctions on neural activity during learning or leveraging what has been learned. Surely, research along those lines is indicated, but the discovery was just published.

Benefit of the Interdisciplinary Approach to AI

That those who reference papers like this can see value in the unification or at least alignment of knowledge across disciplines is most likely beneficial to AI progress and progress in the other fields of cognitive science, bioinformatics, business automation, manufacturing and consumer robotics, psychology, and even law, ethics and philosophy.

That such interest in aligning understanding along interdisciplinary lines is present in AI Stack Exchange is certainly beneficial to the community growth in both professional and social dimensions.

Disparity Between What Works

In the human brain, neurons work. Whether Rosehip neurons are a prerequisite to language, the building of and leveraging of complex models, or transcendent emotions such as love in homo sapiens is unknown and will remain so in the near future. However, we have a fifty millennia long proof of concept.

We also know that artificial networks work. We use them in business, finance, industry, consumer products, and a variety of web services today. When a pop-up asks whether the answer given was helpful, our answer becomes a label in a set of real data from which samples are extracted for machine learning.

Nonetheless, the cells that are working are offspring of the 1957 perceptron with the addition of the application of gradient descent using an efficient corrective signal distribution strategy we call back propagation. The comprehension of neuron function in 1957 was grossly short of what we now know to be functional features of mammalian brain neurons. The Rosehip discovery may widen that gap.

Spiking Networks

The spiking network research more realistically models neurons, and neuromorphic research and development has been placing improved models into VLSI chips. The joint venture between IBM and MIT is another.

Correlating Neural Function to Brain Function

The relationship intelligence and the number of proteins or molecules may not be the most telling. These are more likely relationships between metrics and features and the intelligence of the system.

  • Genetic features that have been identified (22 of them) that directly affect intelligence testing results — For instance the correlation between polymorphisms of the oxytocin receptor genes OXTR rs53576, rs2254298, and rs2228485 and intelligence is known — See the question containing references to discovery of 22 genes that affect intelligence test results significantly
  • Neurochemical expression resulting from environmental factors varying the levels of oxytosin, dopamine, serotonin, neuropeptide Y, and canabinoids which is involved in global and regional functional behavior in the human brain
  • Signal topology (distinct from sizes and counts and distinct from the topology of created by packaging neural nets in the cranial region) — Signal topology is now being identified. Scanning technology has developed to the point where signal paths can be identified by tracking pulses in temporal space and determine causality.
  • Synaptic plasticity, a type of neural plasticity
  • Total number of neurons applied to a particular brain function
  • Impact on axon and cell body thermodynamics on signal transmission, a key element in modelling a brain neuron

None of these are yet modelled in such a way that simulation accuracy has been confirmed, but the need to research along these lines is clearly indicated as this question implies.


It looks like you really have two questions here. I'll try to answer the first one, and you should think about making a separate question for the second.

There is research into using simulated models of biologically realistic neurons. While there are large projects like the Human Brain Project aimed at simulating human brains, there is also a lot of lower-level AI research. SPAWN is an interesting system that got a lot of press a few years ago, and has continued to be developed since then. It uses realistic neurons to simulate several brain-regions at once, creating a surprisingly general AI system that could perform many types of motor and vision tasks using the same basic design.

  • $\begingroup$ The complexity of a neuron cell is too much to be of any use by modern computers. I think most of the projects are replicating the i/p, o/p function rather than actually imitating a neuron.. $\endgroup$
    – user9947
    Commented Sep 10, 2018 at 11:20
  • $\begingroup$ @DuttA, there's a spectrum. See Eliasmith's book Neural Engineering for more. Basically, you call build a far more accurate simulation of a neuron than the usual RLU or Sigmoid models without doing the full biochemical simulation. These more realistic simulations are useful with modern hardware, and can be simulated in very large numbers. $\endgroup$ Commented Sep 10, 2018 at 13:07

It is true that the current Machine learning is based on treating neurons as a component in the whole complexity , mesh of neurons. The focus is more on the architecture rather than understanding or imitating the basic block of it more clearly , i.e. the neurons.

Anirban Bandhopadhyay is a biologist and Neurologist who has studied how the harmony changes the memory element and decision making power in microtubles inside the neurons.

Here, is the snippet of him explaining , and trying to see what exactly computation is , and how the brain does computation.

How does the Brain Act?


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .