Fixing the problems of deep neural networks will require better training data and learning algorithms

Bowers, Jeffrey S.Malhotra, GauravDujmovic, MarinMontero, Milton LleraTsvetkov, ChristianBiscione, ValerioPuebla, GuillermoAdolfi, FedericoHummel, John E.Heaton, Rachel F.Evans, Benjamin D.Mitchell, JeffreyBlything, RyanAnderson, Barton L.Storrs, Katherine R.Fleming, Roland W.Bever, Thomas G.Chomsky, NoamFong, SandiwayPiattelli-Palmarini, MassimoChandran, Keerthi S.Paul, Amrita MukherjeePaul, AvijitGhosh, Kuntalde Vries, Jelmer PhilipFlachot, AlbanMorimoto, TakumaGegenfurtner, Karl R.DiCarlo, James J.Yamins, Daniel L. K.Ferguson, Michael E.Fedorenko, EvelinaBethge, MatthiasBonnen, TylerSchrimpf, MartinGerman, Joseph ScottJacobs, Robert A.Golan, TalTaylor, JohnMarkSchutt, HeikoPeters, BenjaminSommers, Rowan P.Seeliger, KatjaDoerig, AdrienLinton, PaulKonkle, Taliavan Gerven, MarcelKording, KonradRichards, BlakeKietzmann, Tim C.Lindsay, Grace W.Kriegeskorte, NikolausGur, MosheHermann, KatherineNayebi, Aranvan Steenkiste, SjoerdJones, MattHoughton, ConorKazanina, NinaSukumaran, PriyankaKellman, Philip J.Baker, NicholasGarrigan, PatrickPhillips, AustinLu, HongjingKoculak, MarcinWierzchon, MichalLi, Aedan Y.Mur, MariekeLin, HauseLinsley, DrewSerre, ThomasLiu, JianghaoBartolomeo, PaoloLove, Bradley C.Mok, Robert M.Moldoveanu, MihneaOp de Beeck, HansBracci, StefaniaRothkopf, ConstantinBremmer, FrankFiehler, KatjaDobs, KatharinaTriesch, JochenSlagter, Heleen A.Spratling, Michael W.Srivastava, NisheethSifar, AnjaliSrinivasan, NarayananSummerfield, ChristopherThompson, Jessica A. F.Tarr, Michael J.Veit, WalterBrowning, HeatherWichmann, Felix A.Kornblith, SimonGeirhos, RobertXu, YaodaVaziri-Pashkam, MaryamYovel, GalitAbudarham, Naphtali2024-02-212024-02-212024-02-212023-12-0610.1017/S0140525X23001589https://infoscience.epfl.ch/handle/20.500.14299/204956WOS:001127780700001Bowers et al. argue that deep neural networks (DNNs) are poor models of biological vision because they often learn to rival human accuracy by relying on strategies that differ markedly from those of humans. We show that this problem is worsening as DNNs are becoming larger-scale and increasingly more accurate, and prescribe methods for building DNNs that can reliably model biological vision.Life Sciences & BiomedicineFixing the problems of deep neural networks will require better training data and learning algorithmstext::journal::journal article::research article