000917550 001__ 917550
000917550 005__ 20240712112853.0
000917550 0247_ $$2doi$$a10.48550/ARXIV.2211.12386
000917550 0247_ $$2Handle$$a2128/33638
000917550 037__ $$aFZJ-2023-00752
000917550 1001_ $$0P:(DE-Juel1)180221$$aDoncevic, Danimir$$b0$$ufzj
000917550 245__ $$aA Recursively Recurrent Neural Network (R2N2) Architecture for Learning Iterative Algorithms
000917550 260__ $$barXiv$$c2022
000917550 3367_ $$0PUB:(DE-HGF)25$$2PUB:(DE-HGF)$$aPreprint$$bpreprint$$mpreprint$$s1673944665_28073
000917550 3367_ $$2ORCID$$aWORKING_PAPER
000917550 3367_ $$028$$2EndNote$$aElectronic Article
000917550 3367_ $$2DRIVER$$apreprint
000917550 3367_ $$2BibTeX$$aARTICLE
000917550 3367_ $$2DataCite$$aOutput Types/Working Paper
000917550 520__ $$aMeta-learning of numerical algorithms for a given task consist of the data-driven identification and adaptation of an algorithmic structure and the associated hyperparameters. To limit the complexity of the meta-learning problem, neural architectures with a certain inductive bias towards favorable algorithmic structures can, and should, be used. We generalize our previously introduced Runge-Kutta neural network to a recursively recurrent neural network (R2N2) superstructure for the design of customized iterative algorithms. In contrast to off-the-shelf deep learning approaches, it features a distinct division into modules for generation of information and for the subsequent assembly of this information towards a solution. Local information in the form of a subspace is generated by subordinate, inner, iterations of recurrent function evaluations starting at the current outer iterate. The update to the next outer iterate is computed as a linear combination of these evaluations, reducing the residual in this space, and constitutes the output of the network. We demonstrate that regular training of the weight parameters inside the proposed superstructure on input/output data of various computational problem classes yields iterations similar to Krylov solvers for linear equation systems, Newton-Krylov solvers for nonlinear equation systems, and Runge-Kutta integrators for ordinary differential equations. Due to its modularity, the superstructure can be readily extended with functionalities needed to represent more general classes of iterative algorithms traditionally based on Taylor series expansions.
000917550 536__ $$0G:(DE-HGF)POF4-1121$$a1121 - Digitalization and Systems Technology for Flexibility Solutions (POF4-112)$$cPOF4-112$$fPOF IV$$x0
000917550 536__ $$0G:(DE-Juel1)HDS-LEE-20190612$$aHDS LEE - Helmholtz School for Data Science in Life, Earth and Energy (HDS LEE) (HDS-LEE-20190612)$$cHDS-LEE-20190612$$x1
000917550 588__ $$aDataset connected to DataCite
000917550 650_7 $$2Other$$aMachine Learning (cs.LG)
000917550 650_7 $$2Other$$aNumerical Analysis (math.NA)
000917550 650_7 $$2Other$$aFOS: Computer and information sciences
000917550 650_7 $$2Other$$aFOS: Mathematics
000917550 7001_ $$0P:(DE-Juel1)172025$$aMitsos, Alexander$$b1$$ufzj
000917550 7001_ $$0P:(DE-HGF)0$$aGuo, Yue$$b2
000917550 7001_ $$0P:(DE-HGF)0$$aLi, Qianxiao$$b3
000917550 7001_ $$0P:(DE-HGF)0$$aDietrich, Felix$$b4
000917550 7001_ $$0P:(DE-Juel1)172097$$aDahmen, Manuel$$b5$$eCorresponding author$$ufzj
000917550 7001_ $$0P:(DE-HGF)0$$aKevrekidis, Ioannis G.$$b6$$eCorresponding author
000917550 773__ $$a10.48550/ARXIV.2211.12386
000917550 8564_ $$uhttps://juser.fz-juelich.de/record/917550/files/2211.12386.pdf$$yOpenAccess
000917550 909CO $$ooai:juser.fz-juelich.de:917550$$pdnbdelivery$$pdriver$$pVDB$$popen_access$$popenaire
000917550 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)180221$$aForschungszentrum Jülich$$b0$$kFZJ
000917550 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)172025$$aForschungszentrum Jülich$$b1$$kFZJ
000917550 9101_ $$0I:(DE-588b)36225-6$$6P:(DE-Juel1)172025$$aRWTH Aachen$$b1$$kRWTH
000917550 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)172097$$aForschungszentrum Jülich$$b5$$kFZJ
000917550 9131_ $$0G:(DE-HGF)POF4-112$$1G:(DE-HGF)POF4-110$$2G:(DE-HGF)POF4-100$$3G:(DE-HGF)POF4$$4G:(DE-HGF)POF$$9G:(DE-HGF)POF4-1121$$aDE-HGF$$bForschungsbereich Energie$$lEnergiesystemdesign (ESD)$$vDigitalisierung und Systemtechnik$$x0
000917550 9141_ $$y2022
000917550 915__ $$0StatID:(DE-HGF)0510$$2StatID$$aOpenAccess
000917550 920__ $$lyes
000917550 9201_ $$0I:(DE-Juel1)IEK-10-20170217$$kIEK-10$$lModellierung von Energiesystemen$$x0
000917550 9801_ $$aFullTexts
000917550 980__ $$apreprint
000917550 980__ $$aVDB
000917550 980__ $$aUNRESTRICTED
000917550 980__ $$aI:(DE-Juel1)IEK-10-20170217
000917550 981__ $$aI:(DE-Juel1)ICE-1-20170217