Abstract
While most models of randomly connected neural networks assume single-neuron models with simple dynamics, neurons in the brain exhibit complex intrinsic dynamics over multiple timescales. We analyze how the dynamical properties of single neurons and recurrent connections interact to shape the effective dynamics in large randomly connected networks. A novel dynamical mean-field theory for strongly connected networks of multi-dimensional rate neurons shows that the power spectrum of the network activity in the chaotic phase emerges from a nonlinear sharpening of the frequency response function of single neurons. For the case of two-dimensional rate neurons with strong adaptation, we find that the network exhibits a state of "resonant chaos", characterized by robust, narrow-band stochastic oscillations. The coherence of stochastic oscillations is maximal at the onset of chaos and their correlation time scales with the adaptation timescale of single units. Surprisingly, the resonance frequency can be predicted from the properties of isolated neurons, even in the presence of heterogeneity in the adaptation parameters. In the presence of these internally-generated chaotic fluctuations, the transmission of weak, low-frequency signals is strongly enhanced by adaptation, whereas signal transmission is not influenced by adaptation in the non-chaotic regime. Our theoretical framework can be applied to other mechanisms at the level of single neurons, such as synaptic filtering, refractoriness or spike synchronization. These results advance our understanding of the interaction between the dynamics of single units and recurrent connectivity, which is a fundamental step toward the description of biologically realistic neural networks.
Introduction
In theoretical neuroscience, the brain is frequently modeled as a massive, randomly connected network of simple units. While these models successfully replicate the chaotic, unpredictable activity observed in the cortex, they often treat neurons as one-dimensional points lacking internal memory. However, biological neurons are complex entities exhibiting Spike-Frequency Adaptation (SFA), where firing rates gradually decrease under constant stimulation. This study addresses a fundamental question: how do these internal, multi-dimensional properties of individual neurons transform the collective behavior and signal-processing capabilities of large-scale networks?
Methodology
The researchers extended the powerful mathematical framework of Dynamical Mean-Field Theory (DMFT) to handle multi-dimensional rate neurons. Unlike traditional DMFT, which is often limited to one-dimensional units, this new approach allows for the description of neurons with multiple internal variables, such as activation and adaptation.
Key methodological innovations include:
- Multi-dimensional DMFT: A framework developed to describe the average behavior of units embedded in random connections.
- Frequency-Domain Analysis: Instead of solving differential equations in time, the authors utilized an iterative frequency-domain approach. They demonstrated that the network's activity spectrum results from the repeated "sharpening" of a single neuron's frequency response.
- Microscopic Simulations: The theoretical findings were validated through numerical simulations of networks containing 1,000 to 2,000 neurons, with synaptic weights sampled from Gaussian distributions.
Key Results and Insights
Resonant Chaos
The study identifies a novel dynamical regime termed "resonant chaos." In standard models, chaos is typically broad-band, resembling white noise. By incorporating adaptation, the authors found that chaos becomes narrow-band; the network effectively "hums" at a specific frequency. Remarkably, this resonance frequency can be predicted almost perfectly from the properties of a single, isolated neuron. The recurrent connections in the network act as a magnifying glass, sharpening the intrinsic frequency preference of individual units into a dominant peak in the power spectrum.
Enhanced Signal Transmission
One of the most significant findings is the "SNR Paradox." In non-chaotic networks, adaptation does not improve signal transmission because it dampens both the signal and the noise. However, in the chaotic regime, adaptation specifically shapes the internally generated noise. This allows low-frequency signals to be transmitted more effectively, significantly improving the Signal-to-Noise Ratio (SNR). This suggests that resonant chaos acts as a specialized filter for neural communication.
Robustness and Stability
Adaptation serves to stabilize the network, requiring significantly stronger connection strengths to trigger chaos compared to non-adapting models. Furthermore, the resonant chaos state is robust to heterogeneity; even when neurons have widely varying adaptation parameters (quenched disorder), the narrow-band oscillations persist.
Discussion and Limitations
While this framework provides deep insights into the interaction between single-unit dynamics and recurrent connectivity, it relies on rate models rather than spiking models. Spiking neurons introduce discrete electrical events that may lead to additional noise or synchronization effects not captured here. Additionally, the assumption of purely random connectivity differs from the highly structured architecture of biological brains.
Future research could apply this multi-dimensional DMFT framework to other biological mechanisms, such as synaptic filtering, refractoriness, or complex dendritic processing, further bridging the gap between cellular biology and systems-level neuroscience.
Summary updated