Optical Neural Networks: An Energy-Efficient AI Alternative
• Recent research shows promise for optical neural networks as an energy-efficient AI alternative to traditional digital neural nets.
• Novel optical architecture uses wavefront shaping in multimode fibers to enable complex nonlinear optical computation with microwatts of power.
• With few programmable parameters, optical systems can match performance of digital nets with 100x more parameters.
• Precise pulse control in fibers allows efficient nonlinear phenomena programming for optical machine learning.
• Optical neural nets offer a potential solution to unsustainable demands of larger AI models.