Skip to content

Institute Colloquium



Learning at the edge of chaos:
The inner link between complexity, nonlinear waves, and neuromorphic computing

March 22, 2023

2:00 p.m. // Conference room at Leibniz IPHT

Giulia Marcucci obtained a Bachelor’s degree in Mathematics in 2011, a Bachelor’s degree in Physics in 2014, graduated cum laude in Physics in 2016, and defended her Ph.D. thesis in February 2020, under the supervision of Prof. C. Conti at Sapienza University (Rome, Italy). In the same year, she moved to the University of Ottawa (Ontario, Canada) to be a Postdoctoral Fellow in the Quantum Photonics CERC Group led by Prof. R. W. Boyd. From 2021 to 2022, she led the theoretical research in a scientific startup based in London dedicated to the development of biophotonic computational devices. Currently, she is a research fellow at the University of Glasgow working on neuromorphic computing.

Giulia Marcucci has her research background in classical optical nonlinear waves dynamics and extreme events, and data-driven inference and machine-learning algorithms.
Her current interest is in engineering new optical computational systems based on biological reservoirs to exploit  randomness, multimodal propagation, quantum uncertainties, or extremely nonlinear regimes to solve NP-complete problems or to perform neuromorphic computing.
Nonlinear waves' historical role in developing the science of complexity and their physical feature of being widespread in optics and hydrodynamics establish a bridge between two diverse but fundamental fields: nonlinear physics and computational science. Such a link has been opening an endless number of new research routes. Many relevant results on nonlinear waves in photonics and acoustics have assumed major significance in the foundation of new computing models.

In this seminar, I will first report my work on the control of complex nonlinear regimes through topological invariants in photorefractive crystals. Such analysis represents a groundwork for enabling nonlinear waves to do computing, a feature that arises efficiently only when the wave reservoir is at the edge of chaos. Indeed, when waves are both highly nonlinear and controllable, the wave reservoir has two essential properties: given two distinct but similar inputs, their outputs are always distinguishable but never divergent. To demonstrate it, I will present my last works on the engineering of neuromorphic computers, by wave reservoirs.