001008841 001__ 1008841
001008841 005__ 20240313103125.0
001008841 0247_ $$2datacite_doi$$a10.34734/FZJ-2023-02506
001008841 037__ $$aFZJ-2023-02506
001008841 1001_ $$0P:(DE-Juel1)186881$$aWybo, Willem$$b0$$eCorresponding author$$ufzj
001008841 1112_ $$aCosyne 2023$$cMontreal$$d2023-03-08 - 2023-03-16$$wCanada
001008841 245__ $$aDendritic modulation for multitask representation learning in deep feedforward networks
001008841 260__ $$c2023
001008841 3367_ $$033$$2EndNote$$aConference Paper
001008841 3367_ $$2BibTeX$$aINPROCEEDINGS
001008841 3367_ $$2DRIVER$$aconferenceObject
001008841 3367_ $$2ORCID$$aCONFERENCE_POSTER
001008841 3367_ $$2DataCite$$aOutput Types/Conference Poster
001008841 3367_ $$0PUB:(DE-HGF)24$$2PUB:(DE-HGF)$$aPoster$$bposter$$mposter$$s1689922565_8663$$xAfter Call
001008841 520__ $$aFeedforward sensory processing in the brain is generally construed as proceeding through a hierar- chy of layers, each constructing increasingly abstract and invariant representations of sensory inputs. This interpretation is at odds with the observation that activity in sensory processing layers is heavily modulated by contextual signals, such as cross modal information or internal mental states [1]. While it is tempting to assume that such modulations bias the feedforward processing pathway towards de- tection of relevant input features given a context, this induces a dependence on the contextual state in hidden representations at any given layer. The next processing layer in the hierarchy thus has to be able to extract relevant information for each possible context. For this reason, most machine learning approaches to multitask learning apply task-specific output networks to context-independent representations of the inputs, generated by a shared trunk network.Here, we show that a network motif, where a layer of modulated hidden neurons targets an out- put neuron through task-independent feedforward weights, solves multitask learning problems, and that this network motif can be implemented with biophysically realistic neurons that receive context- modulating synaptic inputs on dendritic branches. The dendritic synapses in this motif evolve ac- cording to a Hebbian plasticity rule modulated by a global error signal. We then embed such a motif in each layer of a deep feedforward network, where it generates task-modulated representations of sensory inputs. To learn feedforward weights to the next layer in the network, we apply a contrastive learning objective that predicts whether representations originate either from different inputs, or from different task-modulations of the same input. This self-supervised approach results in deep represen- tation learning of feedforward weights that accommodate a multitude of contexts, without relying on error backpropagation between layers.
001008841 536__ $$0G:(DE-HGF)POF4-5232$$a5232 - Computational Principles (POF4-523)$$cPOF4-523$$fPOF IV$$x0
001008841 536__ $$0G:(EU-Grant)720270$$aHBP SGA1 - Human Brain Project Specific Grant Agreement 1 (720270)$$c720270$$fH2020-Adhoc-2014-20$$x1
001008841 536__ $$0G:(EU-Grant)785907$$aHBP SGA2 - Human Brain Project Specific Grant Agreement 2 (785907)$$c785907$$fH2020-SGA-FETFLAG-HBP-2017$$x2
001008841 536__ $$0G:(EU-Grant)945539$$aHBP SGA3 - Human Brain Project Specific Grant Agreement 3 (945539)$$c945539$$fH2020-SGA-FETFLAG-HBP-2019$$x3
001008841 536__ $$0G:(DE-Juel-1)PF-JARA-SDS005$$aSDS005 - Towards an integrated data science of complex natural systems (PF-JARA-SDS005)$$cPF-JARA-SDS005$$x4
001008841 536__ $$0G:(DE-82)EXS-SF-neuroIC002$$aneuroIC002 - Recurrence and stochasticity for neuro-inspired computation (EXS-SF-neuroIC002)$$cEXS-SF-neuroIC002$$x5
001008841 7001_ $$0P:(DE-Juel1)192408$$aTran, Viet Anh Khoa$$b1$$ufzj
001008841 7001_ $$0P:(DE-HGF)0$$aTsai, Matthias$$b2
001008841 7001_ $$0P:(DE-HGF)0$$aIlling, Bernd$$b3
001008841 7001_ $$0P:(DE-HGF)0$$aJordan, Jakob$$b4
001008841 7001_ $$0P:(DE-HGF)0$$aSenn, Walter$$b5
001008841 7001_ $$0P:(DE-Juel1)151166$$aMorrison, Abigail$$b6$$ufzj
001008841 8564_ $$uhttps://juser.fz-juelich.de/record/1008841/files/Cosyne_2023_dendritic_modulation_abstract%20%281%29.pdf$$yOpenAccess
001008841 909CO $$ooai:juser.fz-juelich.de:1008841$$pec_fundedresources$$pdriver$$pVDB$$popen_access$$popenaire
001008841 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)186881$$aForschungszentrum Jülich$$b0$$kFZJ
001008841 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)192408$$aForschungszentrum Jülich$$b1$$kFZJ
001008841 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)151166$$aForschungszentrum Jülich$$b6$$kFZJ
001008841 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$$x0
001008841 9141_ $$y2023
001008841 915__ $$0StatID:(DE-HGF)0510$$2StatID$$aOpenAccess
001008841 9201_ $$0I:(DE-Juel1)INM-6-20090406$$kINM-6$$lComputational and Systems Neuroscience$$x0
001008841 9201_ $$0I:(DE-Juel1)IAS-6-20130828$$kIAS-6$$lTheoretical Neuroscience$$x1
001008841 9201_ $$0I:(DE-Juel1)INM-10-20170113$$kINM-10$$lJara-Institut Brain structure-function relationships$$x2
001008841 9801_ $$aFullTexts
001008841 980__ $$aposter
001008841 980__ $$aVDB
001008841 980__ $$aUNRESTRICTED
001008841 980__ $$aI:(DE-Juel1)INM-6-20090406
001008841 980__ $$aI:(DE-Juel1)IAS-6-20130828
001008841 980__ $$aI:(DE-Juel1)INM-10-20170113
001008841 981__ $$aI:(DE-Juel1)IAS-6-20130828