Photonic neural networks(PNNs)of sufficiently large physical dimensions and high operation accuracies are envisaged as ideal candidates for breaking the major bottlenecks in the current artificial intelligence archite...Photonic neural networks(PNNs)of sufficiently large physical dimensions and high operation accuracies are envisaged as ideal candidates for breaking the major bottlenecks in the current artificial intelligence architectures in terms of latency,energy efficiency,and computational power.To achieve this vision,it is of vital importance to scale up the PNNs while simultaneously reducing the high demand on the dimensions required by them.The underlying cause of this strategy is the enormous gap between the scales of photonic and electronic integrated circuits.Here,we demonstrate monolithically integrated optical convolutional processors on thin film lithium niobate(TFLN)that harness inherent parallelism in photonics to enable large-scale programmable convolution kernels and,in turn,greatly reduce the dimensions required by subsequent fully connected layers.Experimental validation achieves high classification accuracies of 96%(86%)on the MNIST(Fashion-MNIST)dataset and 84.6%on the AG News dataset while dramatically reducing the required subsequent fully connected layer dimensions to 196×10(from 784×10)and 175×4(from 800×4),respectively.Furthermore,our devices can be driven by commercial field-programmable gate array systems;a unique advantage in addition to their scalable channel number and kernel size.Our architecture provides a solution to build practical machine learning photonic devices.展开更多
Energy materials play an important role in renewable and green energy technologies.The exploration of new materials,including nanomaterials,is important for breaking through the current bottlenecks of energy density a...Energy materials play an important role in renewable and green energy technologies.The exploration of new materials,including nanomaterials,is important for breaking through the current bottlenecks of energy density and charging rates.However,traditional theoretical computational methods face the dilemma of long research cycles.Machine learning methods have in recent years shown considerable potential for accelerating research efforts.However,most approaches are limited to specific properties of particular devices.In this paper,we propose a forward prediction and screening framework for functional materials,which includes database selection,attributes,descriptors,machine learning models,and prediction and screening.Based on the Materials Project database,auto-encoding methods are employed to generate Coulomb matrices as the input to train the convolutional neural networks,which finally screen 12 lithium-ion,6 zinc-ion,and 8 aluminum-ion battery cathode materials satisfying the criteria from 4,300 materials.The results show that the proposed framework can predict material performance well toward rapid initial screening.The proposed framework can provide a specific and complete working process reference for energy materials design work,contributing to the theoretical foundation for the design of core industrial software for materials engineering.展开更多
基金supported by the National Natural Science Foundation of China (Grant Nos.12192251,12334014,62335019,12134001,1230441812474378)+1 种基金the Quantum Science and Technology National Science and Technology Major Project(Grant No.2021ZD0301403)the Shanghai Municipal Science and Technology Major Project (Grant No.2019SHZDZX01)。
文摘Photonic neural networks(PNNs)of sufficiently large physical dimensions and high operation accuracies are envisaged as ideal candidates for breaking the major bottlenecks in the current artificial intelligence architectures in terms of latency,energy efficiency,and computational power.To achieve this vision,it is of vital importance to scale up the PNNs while simultaneously reducing the high demand on the dimensions required by them.The underlying cause of this strategy is the enormous gap between the scales of photonic and electronic integrated circuits.Here,we demonstrate monolithically integrated optical convolutional processors on thin film lithium niobate(TFLN)that harness inherent parallelism in photonics to enable large-scale programmable convolution kernels and,in turn,greatly reduce the dimensions required by subsequent fully connected layers.Experimental validation achieves high classification accuracies of 96%(86%)on the MNIST(Fashion-MNIST)dataset and 84.6%on the AG News dataset while dramatically reducing the required subsequent fully connected layer dimensions to 196×10(from 784×10)and 175×4(from 800×4),respectively.Furthermore,our devices can be driven by commercial field-programmable gate array systems;a unique advantage in addition to their scalable channel number and kernel size.Our architecture provides a solution to build practical machine learning photonic devices.
基金financially supported by the Defense Industrial Technology Development Program(JCKY2021-601B019).
文摘Energy materials play an important role in renewable and green energy technologies.The exploration of new materials,including nanomaterials,is important for breaking through the current bottlenecks of energy density and charging rates.However,traditional theoretical computational methods face the dilemma of long research cycles.Machine learning methods have in recent years shown considerable potential for accelerating research efforts.However,most approaches are limited to specific properties of particular devices.In this paper,we propose a forward prediction and screening framework for functional materials,which includes database selection,attributes,descriptors,machine learning models,and prediction and screening.Based on the Materials Project database,auto-encoding methods are employed to generate Coulomb matrices as the input to train the convolutional neural networks,which finally screen 12 lithium-ion,6 zinc-ion,and 8 aluminum-ion battery cathode materials satisfying the criteria from 4,300 materials.The results show that the proposed framework can predict material performance well toward rapid initial screening.The proposed framework can provide a specific and complete working process reference for energy materials design work,contributing to the theoretical foundation for the design of core industrial software for materials engineering.
