A team of researchers led by MIT has developed an experimental system that uses light instead of electrons to perform machine-learning computations. In a study published in Nature Photonics, the researchers demonstrate the system’s potential as a more powerful and energy-efficient alternative to current digital computers. The system, which uses hundreds of micron-scale lasers, achieved over a 100-fold improvement in energy efficiency and a 25-fold improvement in compute density compared to state-of-the-art digital computers for machine learning.

The team believes that the new system could enable large-scale optoelectronic processors capable of running machine-learning tasks on small devices such as smartphones and other edge devices. The components of the system can be fabricated using existing processes, making it commercially viable within a few years. The researchers anticipate that this technology could allow for the development of machine-learning models that are currently out of reach due to the limitations of today’s supercomputers.

The current work builds on previous research by the team, who have been making progress toward Optical systems for machine learning over the past few years. The researchers developed a compact architecture based on vertical surface-emitting lasers (VCSELs), which address the challenges faced by existing optical neural networks. The VCSEL arrays used in the system were developed by the Reitzenstein group at Technische Universitat Berlin. The team has filed a patent for their work, which was sponsored by several organizations including the U.S. Army Research Office and the National Science Foundation.

Sources: Nature Photonics

The post New Optical System Could Boost Machine-Learning Power appeared first on TS2 SPACE.