Machine Learning Enabled Adaptive Wireless Power Transmission System for Neuroscience Study.

We propose a novel machine learning (ML) approach for the adaptive control of wireless power transmission systems for neuroscience studies. Recent advances in wireless technologies have led to new tools and techniques for neuroscience research, particularly in the context of techniques for optogenetics. Such tools eliminate the need for a battery or a tether to an external power supply and enable experiments that can examine complex behaviors such as social interactions in ways. However, current strategies for radio frequency power control, even with optimized transmission antenna designs, fail more often than they succeed in three-dimensional cages or complex environments that demand coverage over large areas. Here, we propose a ML-based algorithm that can effectively address these issues. In our proposed algorithm, we use deep convolutional network networks (CNN) to automatically track the movement and predict the posture of a lab animal, based on which the antenna system is dynamically switched to activate the antenna that maximizes the power efficiency. This dramatically improves the volumetric and angular coverage in the cage as well as the efficiency of the overall power transmission system in in vitro and in vivo experiments, which showcase the potential for their widespread use in various neuroscience studies.