储层计算的新机遇与挑战 Emerging opportunities and challenges for the future of reservoir computing.docx

naturecommunications

Perspective/10.1038/s41467-024-45187-1

Emergingopportunitiesandchallengesforthefutureofreservoircomputing

Received:15April2023

Accepted:16January2024

publishedonline:06March2024

Checkforupdates

MinYan1,CanHuang1,PeterBienstman2,PeterTino3,WeiLin4,5JieSun1

Reservoircomputingoriginatesintheearly2000s,thecoreideabeingto

utilizedynamicalsystemsasreservoirs(nonlineargeneralizationsofstandardbases)toadaptivelylearnspatiotemporalfeaturesandhiddenpatternsin

complextimeseries.Showntohavethepotentialofachievinghigher-precisionpredictioninchaoticsystems,thosepioneeringworksledtoagreatamountofinterestandfollow-upsinthecommunityofnonlineardynamicsandcomplexsystems.Tounlockthefullcapabilitiesofreservoircomputingtowardsafast,lightweight,andsigniicantlymoreinterpretablelearningframeworkfor

temporaldynamicalsystems,substantiallymoreresearchisneeded.This

Perspectiveintendstoelucidatetheparallelprogressofmathematicaltheory,algorithmdesignandexperimentalrealizationsofreservoircomputing,andidentifyemergingopportunitiesaswellasexistingchallengesforlarge-scaleindustrialadoptionofreservoircomputing,togetherwithafewideasand

viewpointsonhowsomeofthosechallengesmightberesolvedwithjointeffortsbyacademicandindustrialresearchersacrossmultipledisciplines.

Atthecoreoftoday’stechnologicalchallengesistheabilitytoprocessinformationatmassivelysuperiorspeedandaccuracy.Despitelarge-scalesuccessofdeeplearningapproachesinproducingexcitingnewpossibilities1–7,suchmethodsgenerallyrelyontrainingbigmodelsofneuralnetworksposingseverelimitationsontheirdeploymentinthemostcommonapplications8.Infact,thereisagrowingdemandfordeve