Welcome to LION – where optics meets artificial intelligence to push energy and speed efficiency in the neural network computing. Reservoir computing and extreme learning machines are two remarkable paradigms that harness the power of large, fixed nonlinear operators alongside smaller, tunable processors. While they have shown their potential in specific scenarios, they have yet to reach the performance levels of deep, fully trained multilayer networks.
Our project at LION centers around a common thread: enhancing reservoir computing through additional training, elevating its performance while maintaining a streamlined parameter structure compared to conventional multilayer networks.
Our Goal:
At LION, our mission is to revolutionize the way we harness the potential of machine learning. We are dedicated to programming both linear and non-linear optical transforms within novel optical computing framework using a minimal set of parameters. Our innovative approach aims to develop a robust algorithmic framework along with the hardware that enables complex machine learning tasks, all powered by these programmable transforms.
Team:
 |
 |
 |
| Prof. Christophe Moser | Prof. Sylvain Gigan | Prof. Niao He |
 |
 |
|
| Prof. Daniel Brunner | Prof. Rachel Grange |
Photos:
Presentations:
Kickoff meeting – Grange.
Kickoff meeting – Yildirim.
20240624 meeting – Maeder.
20241118 meeting (Intro slides).
Publications:
Yildirim M., Oguz I., Kaufmann F., Escalé M. R., Grange R., Psaltis D., Moser C., Nonlinear optical feature generator for machine learning, APL Photonics 8, 106104 (2023). doi: 10.1063/5.0158611.
Moser C., Psaltis D., Yildirim M., Dinc N. U., Oguz I., Nonlinear processing with only linear optics (nPOLO), Proceedings Volume PC12901, Complex Light and Optical Forces XVIII; PC129010S (2024). doi: 10.1117/12.3000998.
Yildirim M., Dinc N. U., Oguz I., Psaltis D., Moser C., Optic neural networks using multiple scattering for linear and non-linear transformations, Proceedings Volume PC12903, AI and Optical Data Sciences V; PC129030N (2024). doi: 10.1117/12.3001620.
