000912160 001__ 912160
000912160 005__ 20240313103120.0
000912160 0247_ $$2doi$$a10.48550/ARXIV.2202.04925
000912160 0247_ $$2doi$$a10.48550/arXiv.2202.04925
000912160 0247_ $$2Handle$$a2128/32943
000912160 037__ $$aFZJ-2022-05378
000912160 1001_ $$0P:(DE-Juel1)180150$$aFischer, Kirsten$$b0$$eCorresponding author
000912160 245__ $$aDecomposing neural networks as mappings of correlation functions
000912160 260__ $$barXiv$$c2022
000912160 3367_ $$0PUB:(DE-HGF)25$$2PUB:(DE-HGF)$$aPreprint$$bpreprint$$mpreprint$$s1670304027_9967
000912160 3367_ $$2ORCID$$aWORKING_PAPER
000912160 3367_ $$028$$2EndNote$$aElectronic Article
000912160 3367_ $$2DRIVER$$apreprint
000912160 3367_ $$2BibTeX$$aARTICLE
000912160 3367_ $$2DataCite$$aOutput Types/Working Paper
000912160 520__ $$aUnderstanding the functional principles of information processing in deep neural networks continues to be a challenge, in particular for networks with trained and thus non-random weights. To address this issue, we study the mapping between probability distributions implemented by a deep feed-forward network. We characterize this mapping as an iterated transformation of distributions, where the non-linearity in each layer transfers information between different orders of correlation functions. This allows us to identify essential statistics in the data, as well as different information representations that can be used by neural networks. Applied to an XOR task and to MNIST, we show that correlations up to second order predominantly capture the information processing in the internal layers, while the input layer also extracts higher-order correlations from the data. This analysis provides a quantitative and explainable perspective on classification.
000912160 536__ $$0G:(DE-HGF)POF4-5232$$a5232 - Computational Principles (POF4-523)$$cPOF4-523$$fPOF IV$$x0
000912160 536__ $$0G:(DE-HGF)POF4-5234$$a5234 - Emerging NC Architectures (POF4-523)$$cPOF4-523$$fPOF IV$$x1
000912160 536__ $$0G:(DE-HGF)SO-092$$aACA - Advanced Computing Architectures (SO-092)$$cSO-092$$x2
000912160 536__ $$0G:(DE-Juel-1)BMBF-01IS19077A$$aRenormalizedFlows - Transparent Deep Learning with Renormalized Flows (BMBF-01IS19077A)$$cBMBF-01IS19077A$$x3
000912160 536__ $$0G:(DE-Juel1)HGF-SMHB-2014-2018$$aMSNN - Theory of multi-scale neuronal networks (HGF-SMHB-2014-2018)$$cHGF-SMHB-2014-2018$$fMSNN$$x4
000912160 536__ $$0G:(DE-82)EXS-SF-neuroIC002$$aneuroIC002 - Recurrence and stochasticity for neuro-inspired computation (EXS-SF-neuroIC002)$$cEXS-SF-neuroIC002$$x5
000912160 588__ $$aDataset connected to DataCite
000912160 650_7 $$2Other$$aDisordered Systems and Neural Networks (cond-mat.dis-nn)
000912160 650_7 $$2Other$$aMachine Learning (stat.ML)
000912160 650_7 $$2Other$$aFOS: Physical sciences
000912160 650_7 $$2Other$$aFOS: Computer and information sciences
000912160 7001_ $$0P:(DE-Juel1)178936$$aRené, Alexandre$$b1$$ufzj
000912160 7001_ $$0P:(DE-Juel1)171384$$aKeup, Christian$$b2$$ufzj
000912160 7001_ $$0P:(DE-Juel1)174497$$aLayer, Moritz$$b3$$ufzj
000912160 7001_ $$0P:(DE-Juel1)156459$$aDahmen, David$$b4$$ufzj
000912160 7001_ $$0P:(DE-Juel1)144806$$aHelias, Moritz$$b5$$ufzj
000912160 773__ $$a10.48550/arXiv.2202.04925
000912160 8564_ $$uhttps://arxiv.org/abs/2202.04925
000912160 8564_ $$uhttps://juser.fz-juelich.de/record/912160/files/Fischer_2022_arxiv.pdf$$yOpenAccess
000912160 909CO $$ooai:juser.fz-juelich.de:912160$$pdnbdelivery$$pdriver$$pVDB$$popen_access$$popenaire
000912160 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)180150$$aForschungszentrum Jülich$$b0$$kFZJ
000912160 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)178936$$aForschungszentrum Jülich$$b1$$kFZJ
000912160 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)171384$$aForschungszentrum Jülich$$b2$$kFZJ
000912160 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)174497$$aForschungszentrum Jülich$$b3$$kFZJ
000912160 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)156459$$aForschungszentrum Jülich$$b4$$kFZJ
000912160 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)144806$$aForschungszentrum Jülich$$b5$$kFZJ
000912160 9131_ $$0G:(DE-HGF)POF4-523$$1G:(DE-HGF)POF4-520$$2G:(DE-HGF)POF4-500$$3G:(DE-HGF)POF4$$4G:(DE-HGF)POF$$9G:(DE-HGF)POF4-5232$$aDE-HGF$$bKey Technologies$$lNatural, Artificial and Cognitive Information Processing$$vNeuromorphic Computing and Network Dynamics$$x0
000912160 9131_ $$0G:(DE-HGF)POF4-523$$1G:(DE-HGF)POF4-520$$2G:(DE-HGF)POF4-500$$3G:(DE-HGF)POF4$$4G:(DE-HGF)POF$$9G:(DE-HGF)POF4-5234$$aDE-HGF$$bKey Technologies$$lNatural, Artificial and Cognitive Information Processing$$vNeuromorphic Computing and Network Dynamics$$x1
000912160 9141_ $$y2022
000912160 915__ $$0StatID:(DE-HGF)0510$$2StatID$$aOpenAccess
000912160 920__ $$lyes
000912160 9201_ $$0I:(DE-Juel1)INM-6-20090406$$kINM-6$$lComputational and Systems Neuroscience$$x0
000912160 9201_ $$0I:(DE-Juel1)IAS-6-20130828$$kIAS-6$$lTheoretical Neuroscience$$x1
000912160 9201_ $$0I:(DE-Juel1)INM-10-20170113$$kINM-10$$lJara-Institut Brain structure-function relationships$$x2
000912160 9801_ $$aFullTexts
000912160 980__ $$apreprint
000912160 980__ $$aVDB
000912160 980__ $$aUNRESTRICTED
000912160 980__ $$aI:(DE-Juel1)INM-6-20090406
000912160 980__ $$aI:(DE-Juel1)IAS-6-20130828
000912160 980__ $$aI:(DE-Juel1)INM-10-20170113
000912160 981__ $$aI:(DE-Juel1)IAS-6-20130828