Neural networks based on optical and wave physics
Shanhui Fan, Stanford University
Optics and wave physics offer significant advantages in the implementation of neural network algorithms. In particular, matrix-vector multiplication, which represent the most computationally intensive steps of deep neural network algorithms, can be implemented in optics with higher speed and lower energy consumption as compared with digital computations. In this talk, we discuss some of our efforts seeking to advance neural network computing, and to exploit wave physics for recurrent neural network.