<ul><li>Researchers implement a versatile and scalable training algorithm, direct feedback alignment, on a hybrid electronic-photonic platform.</li><li>An optical processing unit achieves large-scale random matrix multiplications at speeds up to 1500 TeraOPS.</li><li>Optical training of modern deep learning architectures, including Transformers with over 1 billion parameters, is performed with good performance on various tasks.</li><li>The hybrid opto-electronic approach shows potential for ultra-deep and wide neural networks, offering a promising way to extend the growth of artificial intelligence beyond traditional approaches.</li></ul>

Streamlined optical training of large-scale modern deep learning architectures with direct feedback alignment

Discover more