- This event has passed.
Physics Colloquium Series: Professor Thomas Carroll
(Physics Lecture Series) Center for Computational Materials Science, Naval Research Laboratory (NRL)
“Nonlinear Dynamics of Reservoir Computing”
A reservoir computer is a relatively new development in the field of neuromorphic computing. In conventional neural networks, such as used for deep learning, training the computer to solve new problems requires the connections between nonlinear nodes in the neural network to be adjusted. In a reservoir computer, the internal network is fixed and the computer is trained by doing a linear fit of the node signals to a training signal. Because of this simplicity, reservoir computers are fast to train and may be built as analog computers. The advantage of the analog system is that it can have low size and weight and consume less power than a digital computer.
No one has a good understanding of how reservoir computers work. The reservoir computer is a nonlinear dynamical system, so I use the tools of a field of physics called nonlinear dynamics to try to understand reservoir computers.