News
Today I had the opportunity to give a seminar talk at the Department of Mechanical Engineering Seminar Series, University of New Mexico, on “Reservoir Computing: Toward Brain-Inspired, Energy-Efficient Estimation & Control.”
Modern deep neural networks achieve remarkable performance but demand substantial compute and energy, while the human brain handles complex perception and decision-making on roughly 20 watts. Reservoir computing offers a path between the two: a fixed, randomly connected recurrent network transforms input signals into a rich, high-dimensional representation, and only a simple linear readout is trained, reducing learning to an efficient regression problem.
The talk covered:
- Neuromorphic computing and why energy efficiency matters
- Feed-forward, recurrent, and spiking neural networks
- Reservoir computing and the echo state property
- Classical control theory: LQR, Kalman filter, and LQG
- Closed-form control with spike coding networks
- Our work on RCUKF (Data-Driven Modeling Meets Bayesian Estimation) and Physics-Informed Reservoir Computing (PIRC)
Thanks to my advisor, Prof. Wenbin Wan, for his guidance, and to Dr. Harish P. Cherukuri, Chair of Mechanical Engineering, for hosting the seminar series. Thank you to everyone who attended!