Competition, stability, and functionality in excitatory-inhibitory neural circuits

Simone Betteti, William Retnaraj, A. S. Davydov, Jorge Cortés, Francesco BulloarXiv 2025 · DOI · arXiv

Abstract

Energy-based models have become a central paradigm for understanding computation and stability in both theoretical neuroscience and machine learning. However, the energetic framework typically relies on symmetry in synaptic or weight matrices - a constraint that excludes biologically realistic systems such as excitatory-inhibitory (E-I) networks. When symmetry is relaxed, the classical notion of a global energy landscape fails, leaving the dynamics of asymmetric neural systems conceptually unanchored. In this work, we extend the energetic framework to asymmetric firing rate networks, revealing an underlying game-theoretic structure for the neural dynamics in which each neuron is an agent that seeks to minimize its own energy. In addition, we exploit rigorous stability principles from network theory to study regulation and balancing of neural activity in E-I networks. We combine the novel game-energetic interpretation and the stability results to revisit standard frameworks in theoretical neuroscience, such as the Wilson-Cowan and lateral inhibition models. These insights allow us to study cortical columns of lateral inhibition microcircuits as contrast enhancer - with the ability to selectively sharpen subtle differences in the environment through hierarchical excitation-inhibition interplay. Our results bridge energetic and game-theoretic views of neural computation, offering a pathway toward the systematic engineering of biologically grounded, dynamically stable neural architectures.

Introduction

In theoretical neuroscience, energy-based models like Hopfield networks provide a powerful lens for understanding how neural circuits reach stable decisions or retrieve memories. These models conceptualize computation as a system descending a global "energy landscape" toward a stable minimum. However, this mathematical elegance hinges on the assumption of synaptic symmetry ($W_{ij} = W_{ji}$), which requires that if neuron A excites neuron B, neuron B must reciprocate with identical strength.

This assumption directly contradicts Dale’s Law, a fundamental principle of biology stating that neurons are either strictly excitatory or strictly inhibitory. Modeling realistic Excitatory-Inhibitory (E-I) networks necessitates asymmetric weight matrices. When symmetry is broken, the global energy landscape disappears, making it difficult to predict whether a network will stabilize, oscillate, or descend into chaos. This paper bridges this gap by reframing asymmetric neural dynamics through the lens of multi-agent game theory.

A Game-Theoretic Extension of Energy Models

To address the lack of a global energy function in asymmetric networks, the authors propose a "game-energetic" framework. Instead of a single global landscape, each neuron is assigned its own local energy function. The network's dynamics are then interpreted as a non-cooperative game where each neuron acts as a selfish agent minimizing its individual cost.

Figure 1: Comparison between a global energy landscape (left) and a game-theoretic framework (right) where each neuron minimizes its own energy $E_i$, converging to a Nash Equilibrium.

In this framework, stable network states correspond to Nash Equilibria—points where no single neuron can further reduce its energy by changing its own firing rate. This approach allows researchers to apply the tools of game theory to biologically realistic, asymmetric circuits.

Stability and Biological Realism

Proving that these "neural games" actually converge to stable states requires rigorous mathematical tools. The authors utilize Lyapunov Diagonal Stability (LDS) from control theory to establish conditions under which an asymmetric E-I network is guaranteed to reach a unique, globally asymptotically stable equilibrium.

Figure 2: Schematic of symmetric (left) vs. asymmetric (right) neural interactions. Biological E-I networks violate the symmetry assumption required by classical energy models.

The authors validated these stability theorems through 27,000 independent Monte Carlo simulations of firing-rate networks, confirming global asymptotic stability with a 99% confidence bound. This provides a robust foundation for engineering stable, biologically grounded neural architectures.

Functional Implications: Oscillations and Contrast Enhancement

The game-theoretic perspective offers new insights into classic neuroscience models:

  • Oscillations as Antagonistic Games: In the Wilson-Cowan model, the authors show that excessive self-excitation transforms a cooperative game into a zero-sum antagonistic game. Since such games lack a Nash Equilibrium, the neurons "chase" each other indefinitely, resulting in the limit cycles observed in brain oscillations.
  • Lateral Inhibition and Decision Making: Circuits where multiple excitatory neurons share a single inhibitory mediator naturally form "winner-takes-all" games. The inhibitory neuron enforces categorical decisions by suppressing weaker signals.
  • Hierarchical Contrast Enhancement: By stacking these circuits into layers—mimicking cortical columns—the network can progressively amplify subtle sensory differences. The authors derived a formula for the required cortical depth ($L$) to categorize an input difference $\epsilon$ with precision $\delta$: $L = \lceil 1 + \ln(\epsilon/\delta) / \ln((1-w_{EE})/w_{EE}) \rceil$.

Conclusion and Future Directions

While the current stability proofs assume synaptic homogeneity for tractability, they provide a vital roadmap for future research. The authors suggest that "reverse game theory" (mechanism design) could be used to artificially synthesize neural architectures with specific desired behaviors. By framing neural dynamics as a game, we can move toward a systematic engineering of brain-like computation that respects biological constraints while maintaining mathematical rigor.

Summary updated

Abstract

Energy-based models have become a central paradigm for understanding computation and stability in both theoretical neuroscience and machine learning. However, the energetic framework typically relies on symmetry in synaptic or weight matrices - a constraint that excludes biologically realistic systems such as excitatory-inhibitory (E-I) networks. When symmetry is relaxed, the classical notion of a global energy landscape fails, leaving the dynamics of asymmetric neural systems conceptually unanchored. In this work, we extend the energetic framework to asymmetric firing rate networks, revealing an underlying game-theoretic structure for the neural dynamics in which each neuron is an agent that seeks to minimize its own energy. In addition, we exploit rigorous stability principles from network theory to study regulation and balancing of neural activity in E-I networks. We combine the novel game-energetic interpretation and the stability results to revisit standard frameworks in theoretical neuroscience, such as the Wilson-Cowan and lateral inhibition models. These insights allow us to study cortical columns of lateral inhibition microcircuits as contrast enhancer - with the ability to selectively sharpen subtle differences in the environment through hierarchical excitation-inhibition interplay. Our results bridge energetic and game-theoretic views of neural computation, offering a pathway toward the systematic engineering of biologically grounded, dynamically stable neural architectures.
BibTeX
@misc{betteti2025competition,
  author = {Simone Betteti and William Retnaraj and A. S. Davydov and Jorge Cortés and Francesco Bullo},
  title = {{Competition, stability, and functionality in excitatory-inhibitory neural circuits}},
  year = {2025},
  eprint = {2512.05252},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2512.05252},
  howpublished = {arXiv},
}
Figure