Computing with Dynamical Systems: Theory and Experiments
Collective dynamical systems offer unique opportunities for computing by harnessing the complex interactions of seemingly simple elements, such as oscillators or spike generators. This is possible when such dynamics can be programmed, controlled, and observed. A classic example is a Hopfield network that has been demonstrated to possess properties akin to associative computing. In this talk, I will present some of our recent work as a part of the recently formed E2CDA NRI center: “EXtremely Energy Efficient Collective ELectronics (EXCEL)”, where we are exploring the computing properties of dynamical systems and their realizations using post-CMOS devices. Such dynamical systems can not only allow efficient associative machines, but can also be shown to solve computationally hard problems, such as optimizations (even belonging to the NP class). I will discuss a couple of such approaches both in terms of the underlying theory and prototypical experimental results. Finally, I will discuss our recent work on stochastic sampling machines that leverage bifurcation induced thermal noise of phase transition materials. Arrangements of such neuron-like elements form the backbone of neuro-mimetic spiking networks. I will finally discuss opportunities, potentials and challenges in realizing such computational systems.
|Computing with Dynamical Systems: Theory and Experiments|
Tuesday, April 18, 2017, 4 p.m.–5 p.m. ET
Durham, NC, United States