IEEE Photonics Society

Boston Photonics Society Chapter

Boston Chapter of the IEEE Photonics Society

Machine Learning and Optical Systems
-- The workshop is being postponed to October due to COVID-19 --
PDF

Wednesday, April 8, 15, 22, 29 and May 6, 2020, 7:00-9:30 PM
Located at MIT Lincoln Laboratory – 3 Forbes Road, Lexington, MA, 02420, USA

Wednesday
April 22, 2020
8:15 PM
 

-- The workshop is being postponed due to COVID-19 --: Towards Large-Scale Optical Neural Networks based on Quantum Photoelectric Multiplication"/>Add to Calendar Add to Calendar

Towards Large-Scale Optical Neural Networks based on Quantum Photoelectric Multiplication

Dr. Ryan Hamerly, Massachusetts Institute of Technology, Cambridge, MA

 

Dr. Ryan Hamerly, Massachusetts Institute of Technology, Cambridge, MA

Abstract:  Progress in deep learning has led to a resource crunch where performance is limited by computing power, which is in turn limited by energy consumption. Optics can improve the speed and energy consumption of neural networks, but current schemes suffer from limited connectivity and the relatively large footprint of low-loss nanophotonic devices. We present a novel approach based on homodyne detection that circumvents these limits and is scalable to large (millions of neurons) networks without sacrificing speed (GHz) or energy consumption (sub-fJ/operation). Here the inputs and weights are both encoded optically, allowing the system to be reprogrammed or trained on the fly. Simulations using pre-trained digit-classification models reveal a standard quantum limit (SQL) to energy consumption, in the range of 10-100 zJ/operation, which is below the thermodynamic (Landauer) limit. This architecture will enable a new class of ultra-low-energy processors for deep learning.

 

Biography:  Ryan had many interests when he was young, but when he saw a Tesla coil in action at high school, he knew he wanted to become a physicist.  He taught himself electromagnetism to build his own Tesla coil, but during his studies at Caltech, he veered off into particle physics and general relativity.  In graduate school at Stanford (Mabuchi group), he returned to electromagnetism, pursuing research on quantum feedback control, quantum optics, and nonlinear optics.  After graduating, he spent a gap year working at NII in Tokyo (Yamamoto group) on quantum annealing and optical parametric oscillator networks for combinatorial optimization.  He is presently a research scientist at NTT and visitor at MIT working with Prof. Dirk Englund on integrated photonics and deep learning.

 

Advance registration and fee required (Open to all IEEE members as well as non-members)

$50/$60 (IEEE Member/Non-Member) early registration fee for ten 1-hour talks over five nights; cost includes coffee and cookies each night, as well as downloadable copies of speakers slides. Early registration deadline March 18th, 2020. Post deadline fee $60/$70 (IEEE Member/Non-Member).


For more information on the technical content of the workshop, contact either:
1) Keisuke Kojima, (kojima@merl.com), Chair
2) Ajay Garg, (ajay.sinclair.garg@ieee.org), Co-Chair
3) Dean Tsang, (tsang@ieee.org), Co-Chair
4) Bill Nelson, (w.nelson@ieee.org), Co-Chair