The Neuron as a Direct Data-Driven Controller

Jason Moore,Alexander Genkin,Magnus Tournoy,Joshua Pughe-Sanford,Rob R. de Ruyter van Steveninck,Dmitri B. Chklovskii

2024-01-03

Abstract:In the quest to model neuronal function amidst gaps in physiological data, a promising strategy is to develop a normative theory that interprets neuronal physiology as optimizing a computational objective. This study extends the current normative models, which primarily optimize prediction, by conceptualizing neurons as optimal feedback controllers. We posit that neurons, especially those beyond early sensory areas, act as controllers, steering their environment towards a specific desired state through their output. This environment comprises both synaptically interlinked neurons and external motor sensory feedback loops, enabling neurons to evaluate the effectiveness of their control via synaptic feedback. Utilizing the novel Direct Data-Driven Control (DD-DC) framework, we model neurons as biologically feasible controllers which implicitly identify loop dynamics, infer latent states and optimize control. Our DD-DC neuron model explains various neurophysiological phenomena: the shift from potentiation to depression in Spike-Timing-Dependent Plasticity (STDP) with its asymmetry, the duration and adaptive nature of feedforward and feedback neuronal filters, the imprecision in spike generation under constant stimulation, and the characteristic operational variability and noise in the brain. Our model presents a significant departure from the traditional, feedforward, instant-response McCulloch-Pitts-Rosenblatt neuron, offering a novel and biologically-informed fundamental unit for constructing neural networks.

Neurons and Cognition,Artificial Intelligence,Systems and Control

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to overcome the challenge of insufficient physiological data in neuron function modeling. Specifically, the researchers proposed a novel framework - Direct Data - Driven Control (DD - DC), regarding neurons as optimal feedback controllers rather than traditional feed - forward prediction models. This framework aims to explain how neurons guide the environment to reach specific desired states through their outputs and evaluate the effectiveness of control effects through synaptic feedback.

### Main problems
1. **Insufficient physiological data**: One of the main problems faced by neuroscience research is the high variability of biological neurons, which makes it very difficult to construct a general brain - function model. Parameters such as the ion - channel density of neurons are difficult to quantify, resulting in limited generalization ability of the model.
2. **Limitations of traditional models**: Existing neuron models, such as the McCulloch - Pitts - Rosenblatt model, are overly simplified, lacking internal feedback and time - dynamic characteristics, and are unable to explain many neurophysiological phenomena.
3. **Balance between prediction and control**: Although prediction is an important aspect of neural computation, prediction models alone cannot fully explain certain physiological properties of neurons, especially those in the motor and premotor areas. These neurons not only need to predict future states but also need to influence the external environment through precise control signals.

### Solutions
1. **DD - DC framework**: The researchers proposed a new DD - DC framework, which allows neurons to directly learn control signals from historical observation data without explicitly representing system dynamics and hidden states. This framework can explain multiple neurophysiological phenomena, such as the enhancement / inhibition transition of STDP (synaptic - time - dependent plasticity), the time expansion and adaptability of feed - forward and feedback filters, the imprecision of the spike - generation mechanism under constant stimulation, and the variability in brain operations.
2. **Feedback control**: Regarding neurons as feedback controllers can better explain how neurons influence the environment through their outputs and evaluate control effects through synaptic feedback. This model is not only applicable to neuroscience but can also be applied to machine learning and artificial intelligence, providing a new basic unit for constructing biologically - inspired artificial neural networks.
3. **The role of noise**: The study found that in order to maintain the persistent excitation condition of the system, the brain may deliberately introduce variability and noise. This noise is particularly important under constant input or control conditions and can restore the function of the DD - DC model, enabling it to operate effectively in an online setting.

### Experimental verification
The paper verified the effectiveness of the DD - DC model through numerical simulation and experimental data. For example, by simulating a potentially unstable scalar dynamic system, the researchers showed how the DD - DC LQR controller successfully identified and maintained the optimal value. In addition, by introducing low - variance noise, the researchers restored the function of the DD - DC model in an online setting, verifying the role of noise in maintaining the persistent excitation condition of the system.

In conclusion, by introducing the DD - DC framework, this paper proposes a new method to understand and model neuron functions, especially in the case of insufficient physiological data, providing a more comprehensive and biologically - reasonable neuron model.