And avoiding “catastrophic forgetting” A team of scientists from the Sigma Laboratory, Center for Electronic Engineering at Tsinghua University in Beijing, China, has unveiled a new L2ONN photonic architecture for training artificial neural intelligence (AI). This architecture is designed to ensure continuous learning of AI and avoid “catastrophic forgetting”. Discuss
© Ferra.ru
L2ONN uses the unique properties of light, such as spatial sparsity and multi-spectral parallelism to enable continuous training of neural networks. Unlike existing photonic neural networks, L2ONN is designed based on the physical nature of the interaction of light with matter.
Experimental evaluations have demonstrated that L2ONN has significant capacity and high energy efficiency compared to electronic neural networks. This architecture is capable of solving complex machine learning problems such as image classification, voice recognition and medical diagnostics.
Scientists hope that the development of a photonic architecture for AI training will accelerate the development of more powerful photonic computing and support advanced machine intelligence technologies.