Koç University Researchers Advance Photonic AI for Image Processing, Raising Energy Efficiency Questions
Researchers at Koç University in Istanbul, led by assistant professor Uğur Teğin, have developed two optical systems designed to offload image processing tasks to light. These systems aim to provide a more concrete alternative to traditional electronic accelerators in AI. One system focuses on preserving and optimizing multiple information channels, particularly color images, achieving 98% accuracy on a skin-image benchmark. The other utilizes a fiber laser to perform nonlinear feature transformation, reaching accuracies between 85% and 99% across various datasets. While these innovations demonstrate that optical hardware can perform complex transformations and shape information for easier classification, the research acknowledges that the systems still rely on electronic components for data preparation, conversion, light generation, detection, cooling, and supporting electronics. The primary goal is to shift meaningful processing stages from software into physical optical hardware, rather than replacing t...
