Neuro-evolutionary stochastic architectures in gauge-covariant neural fields
This work addresses architecture search for neural fields in controlled settings, representing an incremental extension of existing gauge-covariant frameworks.
The authors tackled the problem of stochastic architecture search in neural fields by extending a gauge-covariant framework with evolutionary dynamics, finding that only a fully symmetry-constrained version robustly approaches a near-marginal regime and reproduces predicted low-frequency finite-width spectral behavior.
We extend our gauge-covariant stochastic neural-field framework by promoting architecture-level parameters to slow stochastic variables evolving in function space. Our effective theory is formulated in terms of classical commuting fields and provides symmetry-constrained diagnostics of marginality and finite-width effects through the maximal Lyapunov exponent, the amplification factor, and dressed spectral kernels. On top of this dynamics, we introduce a Markovian evolutionary scheme compatible with the local $U(1)$ structure of the effective model. By using a minimal implementation, the genotype is reduced to the weight-variance parameter $σ_w^2$, and the fitness functional combines spectral agreement, marginal stability, and a symmetry-constrained critical anchor. Comparing three evolutionary models, we find that only the fully symmetry-constrained Ginibre $U(1)$ version robustly approaches a narrow near-marginal regime and reproduces the predicted low-frequency finite-width spectral behavior. These results support the use of symmetry-guided effective stability diagnostics as practical principles for stochastic architecture search in controlled settings.