Sketch of a novel approach to a neural model

arXiv:2209.06865 · q-bio.NC, cond-mat.dis-nn, cs.AI, cs.NE, q-bio.MN · Submitted 2022-09-14 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.

Ines: We're looking at the paper: "Sketch of a novel approach to a neural model".

Marcus: Here's the gist of the paper. In this position paper, we present an account of neuroplasticity with respect to cellinternal processing pathways and their relation to membrane and synaptic plasticity<ref:2209.06865#pg2> The neuron-centric model proposes a paradigm switch from a synapse-centric model (each synapse learns independently, based on its history of use) to a neuron-centric model (each neuron uses signal selection for intracellular pathways to express plasticity at the membrane)<ref:2209.06865#pg14> This new approach is necessary because experimental evidence is not well captured by traditional synapse-centric models<ref:2209.06865#pg15> Pruning redundant reasoning tokens The authors argue that traditional synapse-centric, weight-based models of memorization are not sufficient or adequate to capture the complexity of neuroplasticity<ref:2209.06865#pg2> They propose a model where each neuron has a ’vertical’ dimension where internal parameters steer the external membrane- and synapseexpressed parameters<ref:2209.06865#pg14> A neural model consists of (a) expression of parameters at the membrane, in particular dendritic synapses or spines, and axonal boutons (b) internal parameters in the sub-membrane zone and the cytoplasm with its protein signaling network and (c) core parameters in the nucleus for genetic and epigenetic information<ref:2209.06865#pg14> In this neuron-centric model, each node (=neuron) in the horizontal network has its own internal memory<ref:2209.06865#pg14> Neural transmission and information storage are separated, not automatically combined by coupling strength<ref:2209.06865#pg14> There is filtering and selection of signals for storage<ref:2209.06865#pg14> How the training signal is built The authors define external parameters roughly correspond to membrane proteins, such as receptors and ion channels<ref:2209.06865#pg15> These external parameters undergo plasticity on at least three time-scales: fast (milliseconds, (de)sensitization), intermediate (secondsminutes, endocytosis, insertion), and slow (hours, new proteins and morphological features like spines)<ref:2209.06865#pg15> Only the fast responses, which consist of protein conformational changes, do not need the participation of the internal processing system<ref:2209.06865#pg15> Internal parameters only have indirect effects on neural transmission and remain hidden from other neurons<ref:2209.06865#pg15> The internal system itself cannot regulate its own plasticity<ref:2209.06865#pg14> Where the gains come from The internal computational system that filters and temporally re-sorts signals, that combines signals, and stores them, is a programmable and re-programmable system<ref:2209.06865#pg15> The neuron regulates its responses, keeping itself intact as a functioning cell, and adjusting to its position in a network of other neurons<ref:2209.06865#pg15> Long-term potentiation / depression (LTP/LTD) can be seen as a special case of signal-induced long-term membrane plasticity after strong ongoing or temporally structured stimulation, temporal integration for membrane signals, and extended intracellular responses<ref:2209.06865#pg15> The system is versatile enough with its many control structures to perform interesting functions<ref:2209.06865#pg15> Pruning redundant reasoning tokens The authors suggest that the common theme is that signals are ignored or passed on depending on the responsivity of the neuron, as well as the strength of the signal itself<ref:2209.06865#pg15> The common theme is that signals are ignored or passed on depending on the responsivity of the neuron, as well as the strength of the signal itself<ref:2209.06865#pg15> Pruning redundant reasoning tokens The authors state that synaptic weight adjustment models store the history of the use of the individual synapse, no matter whether by an associative or any other update rule (such as backpropagation)<ref:2209.06865#pg15> Synapse-centric theory of memory has a number of unrealistic consequences<ref:2209.06865#pg15> Huge networks with billions of parameters are needed to solve simple problems by storing millions of patterns<ref:2209.06865#pg15> Pruning redundant reasoning tokens The authors conclude that the proposed neuron model will allow for self-programming of neurons, support modularization into ensembles or groups, and instantiate complex models of cognition<ref:2209.06865#pg15> They anticipate a network of processors with their own individual memory, connected by graded sig10s to offer a highly flexible and sophisticated framework suitable for complex computations<ref:2209.06865#pg15> Pruning redundant reasoning tokens The authors state that the neuron is a highly specialized cell type, strongly compartmentalized with extensive axons and dendrites<ref:2209.06865#pg15> The synapse is a highly specialized connective element<ref:2209.06865#pg15> Plasticity has to deal with the problem of stability vs.

Ines: flexibility, such as overwriting information over time<ref:2209.06865#pg14> Pruning redundant reasoning tokens The authors state that synaptic weight adaptation models store the history of the use of the individual synapse, no matter whether by an associative or any other update rule (such as backpropagation)<ref:2209.06865#pg15> Huge networks with billions of parameters are needed to solve simple problems by storing millions of patterns<ref:2209.06865#pg15> Pruning redundant reasoning tokens The authors state that synaptic weight adjustment models store the history of the use of the individual synapse, no matter whether by an associative or any other update rule (such as backpropagation)<ref:2209.06865#pg15> Huge networks with billions of parameters are needed to solve simple problems by storing millions of patterns<ref:2209.06865#pg15> Pruning redundant reasoning tokens The authors state that synaptic weight adjustment models store the history of the use of the individual synapse, no matter whether by an associative or any other update rule (such as backpropagation)<ref:2209.06865#pg15> Huge networks with billions of parameters are needed to solve simple problems by storing millions of patterns<ref:2209.

Paper summary: Ines: So, to recap what we've covered so far, the paper "Sketch of a novel approach to a neural model" is proposing that we need to switch our focus from how synapses learn independently based on history to how individual neurons use internal processing pathways to determine plasticity at the membrane one <ref:2209.06865#pg1,Sketch of a novel approach to a neural model>.

Marcus: Right. The authors are saying that traditional synapse-centric models are inadequate for capturing the full complexity of neuroplasticity because they simplify how transmission links adapt using just short and long term potentiation or depression one <ref:2209.06865#pg1>.

Ines: They introduce a neuron-centric model where each neuron has an internal 'vertical' dimension with parameters that steer the external membrane and synapse expressions, which includes expression at dendritic spines, axonal boutons, and core parameters in the nucleus one <ref:2209.06865#pg1,steer the external membrane and>.

Yuki: I wonder how this handles the different types of data we get from electrophysiology versus molecular biology because those are usually hard to integrate one <ref:2209.06865#pg1>.

Marcus: The paper suggests that neural transmission and information storage aren't just combined by coupling strength; there is filtering and selection of signals happening at the neuron level one <ref:2209.06865#pg1>. This means each node in the network has its own internal memory one <ref:2209.06865#pg1>.

Ines: And they detail how external parameters, like receptors on the membrane, change across three time scales: fast responses in milliseconds, intermediate changes involving protein insertion over seconds or minutes, and slow morphological changes happening over hours one <ref:2209.06865#pg1>.

Yuki: That seems like a much richer description of the temporal dynamics we observe in real biological systems than just looking at synaptic weights one <ref:2209.06865#pg1>. But they say only the fast responses don't need that internal processing system involved at all one <ref:2209.06865#pg1>.

Marcus: They emphasize that internal parameters only have an indirect effect on neural transmission and remain hidden from other neurons, which is a key difference from models where coupling strength automatically dictates the outcome one <ref:2209.06865#pg1>.

Ines: The paper frames long term potentiation or depression as just one specific case of signal-induced long term membrane plasticity after strong stimulation, involving temporal integration and extended intracellular responses one <ref:2209.06865#pg1>. This shows how the neuron-centric view can account for those complex phenomena one <ref:2209.06865#pg1>.

Yuki: So it sounds like the paper is building a framework that acknowledges the intricate interplay between internal signaling, genetic regulation, and physical changes at the synapse one <ref:2209.06865#pg1>. That's what I find compelling.

Marcus: It’s an attempt to move past models that rely on storing huge networks of synaptic weights just to solve simple problems by remembering millions of patterns one <ref:2209.06865#pg1>. They argue this new approach allows for self-programming neurons and modular organization, which could lead to complex cognition models one <ref:2209.06865#pg1>.

Ines: The authors conclude that this neuron model offers a sophisticated framework because it accounts for the filtering and temporal sorting of signals based on the neuron's specific responsivity one <ref:2209.06865#pg1>. It's a move toward understanding plasticity as an internal regulatory process rather than just an external connection strength adjustment one <ref:2209.06865#pg1>.

Conclusion: Ines: So, looking at "Sketch of a novel approach to a neural model," the authors are proposing this fundamental shift from synapse-centric modeling to a neuron-centric one where internal parameters drive membrane and synaptic plasticity one <ref:2209.06865#pg1,Sketch of a novel approach to a neural model>.

Marcus: They argue that this is necessary because traditional weight-based models struggle to capture the true complexity, especially when dealing with various types of cellular changes across different time scales one <ref:2209.06865#pg1>.

Yuki: The authors are essentially suggesting that understanding neuroplasticity requires us to look deeper into the internal machinery—the protein signaling networks and genetic information—that regulate a neuron's ability to respond one <ref:2209.06865#pg1>.

Ines: They propose that this model lets each neuron have its own internal memory, which allows for self-programming and modular organization within a network of processors one <ref:2209.06865#pg1>. This is seen as a way to move toward more sophisticated models of cognition one <ref:2209.06865#pg1>.

Marcus: The paper’s core implication is that we might need to stop thinking about plasticity just as synaptic weight adjustments and start thinking about the neuron itself as an active regulator responding to its environment one <ref:2209.06865#pg1>.

Yuki: It suggests that future research should focus on how these internal parameters are tuned by genetic and epigenetic information to determine the functional behavior of a neuron one <ref:2209.06865#pg1>.

Ines: Exactly. The authors provide a sketch of a model that acknowledges the stability versus flexibility problem in plasticity by incorporating these deep internal controls one <ref:2209.06865#pg1>. It's about understanding how cells manage their own changes over time one <ref:2209.06865#pg1>.

Gabriele Scheler

Carl Correns Foundation for Mathematical Biology

q-bio.NC, cond-mat.dis-nn, cs.AI, cs.NE, q-bio.MN

Submitted: 2022-09-14

Updated: 2026-11-02

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 81/100

The gist: In this position paper, we present an account of neuroplasticity with respect to cellinternal processing pathways and their relation to membrane and synaptic plasticity<ref:2209.06865#pg2> The…

Key concepts

Neuron-Centric Model
This model views a neural network not as independent synapses, but as individual neurons, each possessing internal parameters. These internal parameters steer how the neuron processes incoming signals and ultimately decides what changes occur at its membrane or synapses. This contrasts with older models where plasticity is solely determined by synaptic history.
Internal Parameters
These are the computational systems located within the neuron's sub-membrane zone and cytoplasm, including protein signaling networks. These parameters filter, select, combine, and store signals before they influence external changes like membrane protein expression or synaptic connections. They allow a single neuron to manage complex information processing.
External Parameters
These refer to the physical manifestations of plasticity at the cell's surface, such as receptors and ion channels on dendrites or spines. These parameters are influenced by internal processing but undergo changes across different time scales—fast, intermediate, and slow—representing different biological processes like protein insertion or morphological changes.
Signal Filtering and Selection
The neuron acts as a programmable system that filters incoming signals based on its own responsivity and the signal's strength. It decides which signals to store or pass on. This internal control allows the neuron to regulate its responses, ensuring it maintains function while adapting to its network environment.

Terminology

Summary

In this position paper, we present an account of neuroplasticity with respect to cellinternal processing pathways and their relation to membrane and synaptic plasticity<ref:2209. p. 0

Where the paper says it

Sketch of a novel approach to a neural model Gabriele Scheler1 1Carl Correns Foundation for Mathematical Biology, Mountain View, CA 94040 Abstract In this position paper, we present an account of neuroplasticity with respect to cellinternal processing pathways and their relation to membrane and synaptic plasticity. We think traditional synapse-centric, weight-based models of memorization are not sufficient or adequate to capture the complexity of neuroplasticity.

06865#pg2> The neuron-centric model proposes a paradigm switch from a synapse-centric model (each synapse learns independently, based on its history of use) to a neuron-centric model (each neuron uses signal selection for intracellular pathways to express plasticity at the membrane)<ref:2209. p. 0
Where the paper says it

The adaptation of these transmission links is overly simplified using short-term and long-term potentiation/depression, assuming weight changes according to use of the transmission link. In contrast to this, we propose a paradigm switch from a synapse-centric model (each synapse learns independently, based on its history of use) to a neuron-centric model (each neuron uses signal selection for intracellular pathways to express plasticity at the membrane).

06865#pg14> This new approach is necessary because experimental evidence is not well captured by traditional synapse-centric models<ref:2209. p. 0
Where the paper says it

We present the neuron as a self-programming device, rather than as passively determined by ongoing input. We believe a new approach to neural modeling is necessary, because the experimental evidence is not well captured by traditional synapse-centric models.

06865#pg15>

Pruning redundant reasoning tokens

The authors argue that traditional synapse-centric, weight-based models of memorization are not sufficient or adequate to capture the complexity of neuroplasticity<ref:2209.06865#pg2> They propose a model where each neuron has a ’vertical’ dimension where internal parameters steer the external membrane- and synapseexpressed parameters<ref:2209. p. 0

Where the paper says it

In contrast to this, we propose a paradigm switch from a synapse-centric model (each synapse learns independently, based on its history of use) to a neuron-centric model (each neuron uses signal selection for intracellular pathways to express plasticity at the membrane). Each neuron has a ’vertical’ dimension where internal parameters steer the external membrane- and synapseexpressed parameters.

06865#pg14> A neural model consists of (a) expression of parameters at the membrane, in particular dendritic synapses or spines, and axonal boutons (b) internal parameters in the sub-membrane zone and the cytoplasm with its protein signaling network and (c) core parameters in the nucleus for genetic and epigenetic information<ref:2209. p. 0
Where the paper says it

Each neuron has a ’vertical’ dimension where internal parameters steer the external membrane- and synapseexpressed parameters. A neural model consists of (a) expression of parameters at the membrane, in particular dendritic synapses or spines, and axonal boutons (b) internal parameters in the sub-membrane zone and the cytoplasm with its protein signaling network and (c) core parameters in the nucleus for genetic and epigenetic information.

06865#pg14> In this neuron-centric model, each node (=neuron) in the horizontal network has its own internal memory<ref:2209. p. 0
Where the paper says it

A neural model consists of (a) expression of parameters at the membrane, in particular dendritic synapses or spines, and axonal boutons (b) internal parameters in the sub-membrane zone and the cytoplasm with its protein signaling network and (c) core parameters in the nucleus for genetic and epigenetic information. In a neuron-centric model, each node (=neuron) in the horizontal network has its own internal memory.

06865#pg14> Neural transmission and information storage are separated, not automatically combined by coupling strength<ref:2209. p. 0
Where the paper says it

In a neuron-centric model, each node (=neuron) in the horizontal network has its own internal memory. Neural transmission and information storage are separated, not automatically combined by coupling strength.

06865#pg14> There is filtering and selection of signals for storage<ref:2209. p. 0
Where the paper says it

Neural transmission and information storage are separated, not automatically combined by coupling strength. There is filtering and selection of signals for storage.

06865#pg14>

How the training signal is built

The authors define external parameters roughly correspond to membrane proteins, such as receptors and ion channels<ref:2209.06865#pg15> These external parameters undergo plasticity on at least three time-scales: fast (milliseconds, (de)sensitization), intermediate (secondsminutes, endocytosis, insertion), and slow (hours, new proteins and morphological features like spines)<ref:2209. p. 1

Where the paper says it

Principles for a neuron-centric model External parameters roughly correspond to membrane proteins, such as receptors and ion channels. They undergo plasticity on at least three time-scales: fast (milliseconds, (de)sensitization), intermediate (secondsminutes, endocytosis, insertion), and slow (hours, new proteins and morphological features like spines).

06865#pg15> Only the fast responses, which consist of protein conformational changes, do not need the participation of the internal processing system<ref:2209. p. 1
Where the paper says it

They undergo plasticity on at least three time-scales: fast (milliseconds, (de)sensitization), intermediate (secondsminutes, endocytosis, insertion), and slow (hours, new proteins and morphological features like spines). Only the fast responses, which consist of protein conformational changes, do not need the participation of the internal processing system.

06865#pg15> Internal parameters only have indirect effects on neural transmission and remain hidden from other neurons<ref:2209. p. 1
Where the paper says it

Membrane parameters - receptors and ion channels - influence neural transmission properties, in particular the activation function, which are visible to other neurons. Internal parameters only have indirect effects on neural transmission and remain hidden from other neurons.

06865#pg15> The internal system itself cannot regulate its own plasticity<ref:2209. p. 3
Where the paper says it

availability of calcium-buffering proteins, release from internal stores) which integrates information from internal and external sources. We assume that the internal system itself cannot regulate its own plasticity [43, 44, 45].

06865#pg14>

Where the gains come from

The internal computational system that filters and temporally re-sorts signals, that combines signals, and stores them, is a programmable and re-programmable system<ref:2209.06865#pg15> The neuron regulates its responses, keeping itself intact as a functioning cell, and adjusting to its position in a network of other neurons<ref:2209. p. 3

Where the paper says it

Long-term potentiation / depression (LTP/LTD) can be seen as a special case of signal-induced long-term membrane plasticity, after strong ongoing or temporally structured stimulation, temporal integration for membrane signals, and extended intracellular responses, such as engaging the kinase/phosphatase system [46]. In general, the neuron regulates its responses, keeping itself intact as a functioning cell, and adjusting to its position in a network of other neurons.

06865#pg15> Long-term potentiation / depression (LTP/LTD) can be seen as a special case of signal-induced long-term membrane plasticity after strong ongoing or temporally structured stimulation, temporal integration for membrane signals, and extended intracellular responses<ref:2209. p. 3
Where the paper says it

Under specified conditions, protein abundances in the processing layer or at the membrane change, adapting the neuron to a new environment. Long-term potentiation / depression (LTP/LTD) can be seen as a special case of signal-induced long-term membrane plasticity, after strong ongoing or temporally structured stimulation, temporal integration for membrane signals, and extended intracellular responses, such as engaging the kinase/phosphatase system [46].

06865#pg15> The system is versatile enough with its many control structures to perform interesting functions<ref:2209. p. 8
Where the paper says it

An example for the power of the ’shared workspace’ is shown in Fig. The system is versatile enough with its many control structures to perform interesting functions.

06865#pg15>

The authors suggest that the common theme is that signals are ignored or passed on depending on the responsivity of the neuron, as well as the strength of the signal itself<ref:2209.06865#pg15> The common theme is that signals are ignored or passed on depending on the responsivity of the neuron, as well as the strength of the signal itself<ref:2209. p. 6

Where the paper says it

Experimentally, the effect of responsivity has been studied along these lines by [85, 33, 86]). The common theme is that signals are ignored or passed on depending on the responsivity of the neuron, as well as the strength of the signal itself.

06865#pg15>

The authors state that synaptic weight adjustment models store the history of the use of the individual synapse, no matter whether by an associative or any other update rule (such as backpropagation)<ref:2209.06865#pg15> Synapse-centric theory of memory has a number of unrealistic consequences<ref:2209. p. 0

Where the paper says it

We believe a new approach to neural modeling is necessary, because the experimental evidence is not well captured by traditional synapse-centric models. Ultimately, we are interested in the possibilities of a flexible memory system that processes external signals according to its inherent structure.

06865#pg15> Huge networks with billions of parameters are needed to solve simple problems by storing millions of patterns<ref:2209.06865#pg15>

The authors conclude that the proposed neuron model will allow for self-programming of neurons, support modularization into ensembles or groups, and instantiate complex models of cognition<ref:2209.06865#pg15> They anticipate a network of processors with their own individual memory, connected by graded sig10s to offer a highly flexible and sophisticated framework suitable for complex computations<ref:2209. p. 9

Where the paper says it

In general, lasting plasticity requires this step. Overall, we anticipate a network of processors with their own individual memory, connected by graded sig10

06865#pg15>

The authors state that the neuron is a highly specialized cell type, strongly compartmentalized with extensive axons and dendrites<ref:2209.06865#pg15> The synapse is a highly specialized connective element<ref:2209.06865#pg15> Plasticity has to deal with the problem of stability vs. flexibility, such as overwriting information over time<ref:2209. p. 0

Where the paper says it

Ultimately, we are interested in the possibilities of a flexible memory system that processes external signals according to its inherent structure. Keywords plasticity, learning, neural model, vertical integration, internal parameters, neuroplasticity, memory, synaptic plasticity 1 arXiv:2209.

06865#pg14>

The authors state that synaptic weight adaptation models store the history of the use of the individual synapse, no matter whether by an associative or any other update rule (such as backpropagation)<ref:2209.06865#pg15> Huge networks with billions of parameters are needed to solve simple problems by storing millions of patterns<ref:2209.06865#pg15>

The authors state that synaptic weight adjustment models store the history of the use of the individual synapse, no matter whether by an associative or any other update rule (such as backpropagation)<ref:2209.06865#pg15> Huge networks with billions of parameters are needed to solve simple problems by storing millions of patterns<ref:2209.06865#pg15>

The authors state that synaptic weight adjustment models store the history of the use of the individual synapse, no matter whether by an associative or any other update rule (such as backpropagation)<ref:2209.06865#pg15> Huge networks with billions of parameters are needed to solve simple problems by storing millions of patterns<ref:2209.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Internal parameter modeling for signal selection: The AI system can implement selection of signals by having internal parameters that determine whether a transmission event leaves a trace, as suggested by Not every transmission event leaves a trace.

  2. Neuron-centric processing architecture: The AI system should be structured around neurons acting as self-programming devices, where each node has its own internal memory and uses signal selection for intracellular pathways to express plasticity at the membrane.

  3. Hierarchical information storage and retrieval: The system can utilize long-term memory of a type that synaptic weight models lack by having core parameters in the nucleus for genetic and epigenetic information, enabling write-once memory events.

  4. Dynamic adaptation based on internal state: The AI can exhibit state-dependent processing where the processor itself has state-dependence, meaning signals are transferred to actuators based on whether a main signaling pathway is highly active.

  5. Encoding neuronal identity through neuromodulation: The system can use neuromodulation as a combinatorial code, where Each neuronal type is differentiated by the NM receptors it carries, allowing for constrained inferences based on receptor labels.

Abstract

We present an account of neuroplasticity with respect to cell-internal processing pathways in relation to membrane and synaptic plasticity. We think traditional synapse-centric, weight-based models of memorization are not sufficient or adequate to capture the complexity of neuroplasticity. In these accounts, the model is a network of neurons connected by adaptive transmission links. The adaptation of the transmission links relies on weight changes according to use of the transmission link (short-term and long-term potentiation/depression). In contrast, we propose a paradigm switch from a synapse-centric model (each synapse learns independently, based on its history of use) to a neuron-centric model (each neuron uses signal selection for intracellular pathways to express plasticity at the membrane). A neural model consists of (a) expression of parameters at the membrane, in particular dendritic synapses or spines, and axonal boutons (b) internal parameters in the sub-membrane zone and the cytoplasm with its protein signaling network and (c) core parameters in the nucleus for genetic and epigenetic information. In a neuron-centric model, each node (=neuron) in the network has its own internal memory. Neural transmission and information storage are separated, not automatically combined by coupling strength. There is filtering and selection of signals for storage. Not every transmission event leaves a trace. This represents an important conceptual advance over synaptic weight models. We present the neuron as a self-programming device, rather than as passively determined by ongoing input. We believe a new approach to neural modeling is necessary, because the experimental evidence is not well captured by traditional synapse-centric models. Ultimately, we are interested in the possibilities of a flexible memory system that processes external signals according to its inherent structure.

Sources

Related papers