In the new research, scientists describe their desire to use photons for tensor processing—a form of computer math where entire sets of data are digested together instead of just one datum at a time. This is a paradigm for data crunching in which numbers and other values can be stored across many dimensions in a large math structure. This type of math has potential value in many fields, but for machine learning structures like large language models (LLMs) and other forms of artificial intelligence (AI), it’s considered a gamechanger because it’s very compatible with model training and other AI work.
Inside your traditional computer, everything you see, type, and click on is a high level representation of code. Your hardware has to be ready for you to do a variety of tasks as a general user, from crunching a huge Excel spreadsheet into a marketing report to playing a cutting-edge video game. But AI often does not need breadth in that way. Its hardware can focus on specific types of math and storage that are the most helpful and efficient, and transitioning these processes from electron to photon computing could be even faster.
In their paper, the scientists in China lay out a new photon process where just one light source can do more than one tensor operation at once. This is essential, because this math involves multiplying huge grids of values into other huge grids of values, typically resulting in a solution represented by an even larger grid. Inside the computer, if one math problem can’t run in tandem with other operations, the system can get really bogged down and slow.
Website Link: molecularbiologist.org/
Contact Mail ID : support@molecularbiologist.org
Nomination Link : https://molecularbiologist.org/award-nomination/?ecategory=Awards&rcategory=Awardee
#PhotonicComputing
#OpticalProcessors
#LightBasedChips
#NextGenComputing
#SiliconPhotonics
#OpticalAI
#PhotonicsRevolution Memory full
No comments:
Post a Comment