Artificial neural networks with internal dynamics exhibit remarkable capability in processing information.Reservoir computing(RC)is a canonical example that features rich computing expressivity and compatibility with ...Artificial neural networks with internal dynamics exhibit remarkable capability in processing information.Reservoir computing(RC)is a canonical example that features rich computing expressivity and compatibility with physical implementations for enhanced efficiency.Recently,a new RC paradigm known as next generation reservoir computing(NGRC)further improves expressivity but compromises its physical openness,posing challenges for realizations in physical systems.Here we demonstrate optical NGRC with computations performed by light scattering through disordered media.In contrast to conventional optical RC implementations,we directly and solely drive our optical reservoir with time-delayed inputs.Much like digital NGRC that relies on polynomial features of delayed inputs,our optical reservoir also implicitly generates these polynomial features for desired functionalities.By leveraging the domain knowledge of the reservoir inputs,we show that the optical NGRC not only predicts the short-term dynamics of the low-dimensional Lorenz63 and large-scale Kuramoto-Sivashinsky chaotic time series,but also replicates their long-term ergodic properties.Optical NGRC shows superiority in shorter training length and fewer hyperparameters compared to conventional optical RC based on scattering media,while achieving better forecasting performance.Our optical NGRC framework may inspire the realization of NGRC in other physical RC systems,new applications beyond time-series processing,and the development of deep and parallel architectures broadly.展开更多
In this study,we examined plant health prediction through the Normalized Difference Vegetation Index(NDVI)calculated from satellite image derived reflectance values in the near-infrared and red spectra.The problem is ...In this study,we examined plant health prediction through the Normalized Difference Vegetation Index(NDVI)calculated from satellite image derived reflectance values in the near-infrared and red spectra.The problem is formulated as a temporal data prediction problem.Using MODIS/Terra Vegetation Indices 16-Day L3 Global 250 m SIN Grid V061 dataset,we designed and implemented Reservoir Computing(RC)models and transformer-based models including pretrained language model,and compared the prediction performance of these models to traditional machine learning and deep learning methods such as Nonlinear Regression,Decision Tree,Convolutional Neural Network(CNN),Long Short-Term Memory(LSTM)network,and DLinear.It is observed that the DLinear/LSTM model showed exceptional predictive accuracy,while the pretrained RC model significantly enhanced traditional RC model forecasts.Additionally,Frozen Pretrained Transformer(FPT),a pretrained language model,showed superior performance in predicting specific NDVI values(most often peak or lowest NDVI),suggesting its effectiveness in precise temporal predictions.Furthermore,transformer-based models,specifically PatchTST and FPT,demonstrated substantial mean squared error reductions,particularly in limited data scenarios(1%,5%,15%and 50%sample sizes),indicating their robustness in precise NDVI temporal predictions when data is limited.The findings in this study demonstrated the effectiveness of emerging machine learning techniques such as reservoir computing and pretrained language model for remote sensing and their contributions in precision agriculture.展开更多
基金supported by Swiss National Science Foundation(SNF)projects LION,ERC SMARTIES and Institut Universitaire de France.H.W.acknowledges China Scholarship Council and National Natural Science Foundation of China(623B2064 and 62275137)J.H.acknowledges SNF fellowship(P2ELP2_199825)+3 种基金Y.B.acknowledges the support from Basic Science Research Program through the National Research Foundation of Korea(NRF)funded by the Ministry of Education(2022R1A6A3A03072108)European Union’s Horizon Europe research and innovation program(N.101105899)Q.L.acknowledges National Natural Science Foundation of China(62275137)the Tsinghua University(Department of Precision Instrument)-North Laser Research Institute Co.,Ltd Joint Research Center for Advanced Laser Technology(20244910194).
文摘Artificial neural networks with internal dynamics exhibit remarkable capability in processing information.Reservoir computing(RC)is a canonical example that features rich computing expressivity and compatibility with physical implementations for enhanced efficiency.Recently,a new RC paradigm known as next generation reservoir computing(NGRC)further improves expressivity but compromises its physical openness,posing challenges for realizations in physical systems.Here we demonstrate optical NGRC with computations performed by light scattering through disordered media.In contrast to conventional optical RC implementations,we directly and solely drive our optical reservoir with time-delayed inputs.Much like digital NGRC that relies on polynomial features of delayed inputs,our optical reservoir also implicitly generates these polynomial features for desired functionalities.By leveraging the domain knowledge of the reservoir inputs,we show that the optical NGRC not only predicts the short-term dynamics of the low-dimensional Lorenz63 and large-scale Kuramoto-Sivashinsky chaotic time series,but also replicates their long-term ergodic properties.Optical NGRC shows superiority in shorter training length and fewer hyperparameters compared to conventional optical RC based on scattering media,while achieving better forecasting performance.Our optical NGRC framework may inspire the realization of NGRC in other physical RC systems,new applications beyond time-series processing,and the development of deep and parallel architectures broadly.
基金supported by USDA/NIFA Award 2022-38821-37338(Project No.TEXXQRD9908).
文摘In this study,we examined plant health prediction through the Normalized Difference Vegetation Index(NDVI)calculated from satellite image derived reflectance values in the near-infrared and red spectra.The problem is formulated as a temporal data prediction problem.Using MODIS/Terra Vegetation Indices 16-Day L3 Global 250 m SIN Grid V061 dataset,we designed and implemented Reservoir Computing(RC)models and transformer-based models including pretrained language model,and compared the prediction performance of these models to traditional machine learning and deep learning methods such as Nonlinear Regression,Decision Tree,Convolutional Neural Network(CNN),Long Short-Term Memory(LSTM)network,and DLinear.It is observed that the DLinear/LSTM model showed exceptional predictive accuracy,while the pretrained RC model significantly enhanced traditional RC model forecasts.Additionally,Frozen Pretrained Transformer(FPT),a pretrained language model,showed superior performance in predicting specific NDVI values(most often peak or lowest NDVI),suggesting its effectiveness in precise temporal predictions.Furthermore,transformer-based models,specifically PatchTST and FPT,demonstrated substantial mean squared error reductions,particularly in limited data scenarios(1%,5%,15%and 50%sample sizes),indicating their robustness in precise NDVI temporal predictions when data is limited.The findings in this study demonstrated the effectiveness of emerging machine learning techniques such as reservoir computing and pretrained language model for remote sensing and their contributions in precision agriculture.
