Information-energy tradeoffs in an escape-rate population model
Phys. Rev. E 114, 044129 – Published 19 October, 2026
DOI: https://doi.org/10.1103/nr7w-18m6
Abstract
Efficient coding of dynamical signals requires neural activity to balance information transmission against firing cost on the timescale of the input. We study this balance in a stochastic escape-rate population with membrane filtering, adaptation, and weak mean-field feedback. A cross-validated decoding rate is optimized against mean firing rate and sparse hazard complexity. Across eleven stimulus correlation times, the optimized rate varies nonmonotonically over the tested range rather than following a scale-free law: it is highest for inputs faster than the membrane timescale, reaches a local minimum at the membrane timescale, rebounds into a broad 5–6 Hz plateau, and falls for the slowest input. Switching-variance stimuli produce lower efficiency than stationary Gaussian stimuli and evoke transient rate changes after variance transitions. Sparse quadratic terms extend the low-rate Pareto front mainly through the interaction between membrane state and population activity. Under a matched restart budget, warm-started six-parameter searches reach a higher objective than direct six-parameter searches at nine of eleven input timescales. Null, spike-feature, finite-size, baseline-cost, and estimator checks support the stability of these comparisons. The results show how the relation between stimulus and intrinsic model timescales organizes the information-energy operating point.
