Light-Based Artificial Intelligence Computing Systems Developed at Koç University
To address artificial intelligence's growing need for energy and processing power, two new photonic systems that use light instead of traditional electronic circuits have been developed.
Dr. Uğur Teğin, a faculty member in the Department of Electrical and Electronics Engineering at Koç University, designed two innovative systems utilizing light to ease the burden on artificial intelligence computations.
Artificial Intelligence and the Energy Problem
With the widespread adoption of artificial intelligence technologies, the increasing demand for processing power and energy stands out as a significant challenge on a global scale.
To find a solution to this problem, researchers from Koç University developed new approaches utilizing light instead of traditional electronic circuits.
Programmable Photonic Network
In the first study published in the journal Communications Engineering, a programmable photonic neural network was developed, making it suitable for classifying information in colored images.
In the task of classifying dermatological diseases, it was observed that the accuracy rate increased from 68 percent to 98 percent following optical processing.
Laser-Based Physical Processor
In the second study published in the journal Communications Physics, a multimode fiber laser was used as a physical processor for machine learning.
Following the optical processing procedure, accuracy between 85 percent and 99 percent was achieved on different datasets using a simple classification method.
Sustainable Artificial Intelligence Future
These two studies conducted at Koç University highlight the complementary strengths of photonic computing.
The obtained results lay the groundwork for faster, energy-efficient, and sustainable new-generation artificial intelligence systems.