Hyperparameter optimization for neural-network models in molecular communication

Develop systematic hyperparameter optimization methods for neural-network architectures applied to molecular communication, including the tuning of the number of nodes, hidden layers, learning rate, and related design parameters, and quantify their impact on performance beyond ad hoc manual selection.

Background

The paper notes that current molecular communication studies employing neural networks typically select hyperparameters (e.g., architecture depth, neuron counts, learning rates) manually without principled optimization or analysis of their downstream effects. This undermines reproducibility and may yield suboptimal models, especially in channels with significant inter-symbol interference or time variability.

A dedicated optimization framework tailored to MC tasks would support better model performance and generalization, enabling fair comparisons across architectures and guiding practitioners in resource-constrained nanoscale settings.