Can information transmission be energy-efficient? This research explores the consequences of energy efficiency in information processing, relevant to both biological sensory systems and low-power electronic devices. Two regimes are considered: maximizing information rate under a power constraint ("immediate" regime) and maximizing transmission rate per power cost ("exploratory" regime). In the absence of noise, discrete inputs are optimally encoded into Boltzmann distributed output symbols. The Arimoto-Blahut algorithm, generalized for cost constraints, is used to derive and interpret symbol distribution for energy-efficient coding in noisy channels. The study discusses potential extensions of these results to neurobiological systems, providing insights into the energetic constraints shaping information processing in diverse systems.
Published in Neural Computation, this paper aligns with the journal's focus on computational and theoretical approaches to understanding neural systems. By exploring the principles of energy-efficient information transmission in the context of neural systems and applying computational methods, the study contributes to the journal's scope of bridging neuroscience and computation.