An information theoretic parameter tuning for MEMS-based reservoir computing

With respect to the next frontier of neuromorphic sensing, we propose a parameter tuning method based on mutual information criteria for MEMS-based reservoir computing. It is required MEMS reservoirs tune balance linear and nonlinear characteristics control their dynamical behaviors depending driving forces, such as chaos hysteresis. We focus pre-training machine learning called intrinsic plasticity (IP) learning, apply it controlling reservoirs. First, demonstrate simulation results suppression. Next, applied our IP Finally, show that approach can improve prediction accuracy in transformation tasks.