001025210 001__ 1025210 001025210 005__ 20250203103227.0 001025210 0247_ $$2doi$$a10.29363/nanoge.neumatdecas.2023.013 001025210 037__ $$aFZJ-2024-02778 001025210 1001_ $$0P:(DE-Juel1)188273$$aNeftci, Emre$$b0$$ufzj 001025210 1112_ $$aNeuromorphic Materials, Devices, Circuits and Systems$$cValència$$d2023-01-23 - 2023-01-25$$wSpain 001025210 245__ $$aTraining-to-Learn with Memristive Devices 001025210 260__ $$c2023 001025210 3367_ $$033$$2EndNote$$aConference Paper 001025210 3367_ $$2DataCite$$aOther 001025210 3367_ $$2BibTeX$$aINPROCEEDINGS 001025210 3367_ $$2DRIVER$$aconferenceObject 001025210 3367_ $$2ORCID$$aLECTURE_SPEECH 001025210 3367_ $$0PUB:(DE-HGF)6$$2PUB:(DE-HGF)$$aConference Presentation$$bconf$$mconf$$s1714574803_3667$$xInvited 001025210 520__ $$aMemristive crossbar arrays are promising non-von Neumann computing technologies to enable real-world, onlinelearning in neural networks. However, their deployment to real-world learning problems is hindered by their non-linearitiesin conductance updates, variation during operation, fabrication mismatch and the realities of gradient descent training. In thiswork, we show that, with a phenomenological model of the device and bi-level optimization, it is possible to pre-train the neuralnetwork to be largely insensitive to such non-idealities on learning tasks. We demonstrate this effect using Model Agnostic Meta Learning (MAML) and a differentiable model of the conductance update on the Omniglot few-shot learning task. Since pre-training is a necessary procedure for any on-line learning scenario at the edge, our results may pave the way towards real-world applications of memristive devices without significant adaption overhead. 001025210 536__ $$0G:(DE-HGF)POF4-5234$$a5234 - Emerging NC Architectures (POF4-523)$$cPOF4-523$$fPOF IV$$x0 001025210 588__ $$aDataset connected to CrossRef Conference 001025210 7001_ $$0P:(DE-Juel1)190500$$aYu, Zhenming$$b1$$ufzj 001025210 7001_ $$0P:(DE-Juel1)194421$$aLeroux, Nathan$$b2 001025210 773__ $$a10.29363/nanoge.neumatdecas.2023.013 001025210 8564_ $$uhttps://www.nanoge.org/proceedings/NeuMatDeCaS/636e689d63c3311959c7c0fa 001025210 909CO $$ooai:juser.fz-juelich.de:1025210$$pVDB 001025210 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)188273$$aForschungszentrum Jülich$$b0$$kFZJ 001025210 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)190500$$aForschungszentrum Jülich$$b1$$kFZJ 001025210 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)194421$$aForschungszentrum Jülich$$b2$$kFZJ 001025210 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$$x0 001025210 9141_ $$y2024 001025210 9201_ $$0I:(DE-Juel1)PGI-15-20210701$$kPGI-15$$lNeuromorphic Software Eco System$$x0 001025210 980__ $$aconf 001025210 980__ $$aVDB 001025210 980__ $$aI:(DE-Juel1)PGI-15-20210701 001025210 980__ $$aUNRESTRICTED