Lighting Up the Network
From the discovery of fire to the introduction of the fiber-optic cable, light has a way of being a game-changer in whatever field it is introduced in. The realm of machine learning is no exception. Last month, scientists have devised a way to create neural networks that literally perform at the speed of light.
Researchers at the University of California, Los Angeles (UCLA) have created a set of 3D-printed panels that use light to process information like a traditional neural network. This was achieved by using translucent surfaces that bounced light in specific directions. This then caused light to be refracted to certain panels in the network. By having detectors at the end of this makeshift network, the prototype circuit can make decisions similar to how a traditional neural network obtains a result through the passage of data in its layered “neurons.”
While early results are promising, there are still certain issues that need to be addressed. For instance, the device can only perform basic operations like character recognition, unlike traditional neural networks, which can perform complex tasks like deep learning. Furthermore, the 3D-printed panels used in the design must be manufactured at certain shapes, which is impractical for mass production.
However, the research’s proponents remain optimistic about the implications of their work. After all, their light-based neural network possesses unmatched processing speeds, and requires little power compared to its electronic counterparts. Only time will tell if this radiant innovation will blaze a trail towards a new era of machine learning and artificial intelligence.
Want to know more about this innovation? Just follow this link!
Like us on Facebook!
- Fun Facts and Trivia brought to you by the Research and Development Committee -
Thank you and God Bless!