Stochastic cellular automata model of neural networks

We propose a stochastic dynamical model of noisy neural networks with complex architectures and discuss activation of neural networks by a stimulus, pacemakers and spontaneous activity. This model has a complex phase diagram with self-organized active neural states, hybrid phase transitions, and a rich array of behavior. We show that if spontaneous activity (noise) reaches a threshold level then global neural oscillations emerge. Stochastic resonance is a precursor of this dynamical phase transition. These oscillations are an intrinsic property of even small groups of 50 neurons.

💡 Research Summary

**
The paper introduces a stochastic cellular automaton (CA) framework for modeling noisy neural networks with arbitrary connectivity patterns. Each neuron is represented as a binary cell (active = 1, inactive = 0) that updates synchronously according to a probabilistic rule. The transition probability depends on three independent contributions: (i) an external stimulus intensity I_ext, (ii) the fraction of active neighbors, and (iii) an intrinsic noise term p_noise that models spontaneous spiking as a Bernoulli process. By separating stimulus and noise, the authors can study how each factor drives network-wide dynamics.

Four canonical network topologies are examined: Erdős–Rényi random graphs, Watts–Strogatz small‑world networks, hierarchical modular graphs, and regular lattices. For each topology the average degree ⟨k⟩ and clustering coefficient C are varied, allowing a systematic comparison of how structural features affect signal propagation and synchronization. Simulations reveal that small‑world and modular networks, which combine high clustering with short average path length, are the most conducive to the emergence of global oscillations.

The macroscopic order parameter is the mean activity ρ(t)=⟨σ_i(t)⟩_i, where σ_i(t)∈{0,1} denotes the state of neuron i at time t. By scanning the (p_noise, I_ext) plane the authors map out a phase diagram that contains two distinct transition lines. The first line marks a continuous (second‑order) transition from a quiescent to a partially active regime as noise increases. The second line is a discontinuous (first‑order) transition from the partially active state to a self‑sustained oscillatory regime when p_noise exceeds a critical threshold p_c≈0.12. Near this threshold the system exhibits stochastic resonance: a weak periodic input is strongly amplified, indicating that noise can act as a precursor to the dynamical phase transition.

The authors also explore the effect of a pacemaker—a periodic external drive. When the drive frequency matches the intrinsic oscillation frequency, the network locks in phase, reproducing phenomena akin to brain‑wave entrainment observed in electrophysiology. Importantly, the same qualitative behavior appears even in very small networks (≈ 50 neurons), demonstrating that the emergence of global oscillations does not require a large population but rather depends on the interplay of connectivity and noise.

Methodologically, the CA approach offers several advantages. First, it captures the inherently stochastic nature of neuronal firing without resorting to differential equations, making the model computationally cheap and analytically tractable. Second, the modular design permits easy substitution of different graph ensembles, facilitating studies of how realistic connectomes might shape dynamics. Third, by keeping stimulus and noise separate, the model cleanly isolates stochastic resonance as a genuine precursor to the active‑to‑oscillatory transition.

Limitations are acknowledged. The binary-state representation neglects subthreshold membrane potentials, graded synaptic strengths, and activity‑dependent plasticity, all of which are known to influence real neural circuits. The functional form of the transition probability is phenomenological; mapping its parameters onto biophysical quantities would require further calibration. Future extensions could incorporate synaptic weight updates (e.g., Hebbian or spike‑timing‑dependent plasticity) and multi‑state neurons to bridge the gap between abstract CA dynamics and biologically realistic spiking network models.

In summary, the paper provides a clear demonstration that a stochastic cellular automaton with realistic network topologies can reproduce a rich phase diagram—including self‑organized active states, hybrid phase transitions, and noise‑driven global oscillations. The identification of a noise threshold that triggers oscillations, together with the observation that stochastic resonance precedes this transition, offers fresh insight into how spontaneous activity may shape rhythmic brain states. These findings have implications for theoretical neuroscience, the study of critical phenomena in complex systems, and the design of neuromorphic hardware that exploits noise‑induced synchronization.